Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/page_alloc.c | |
3 | * | |
4 | * Manages the free list, the system allocates free pages here. | |
5 | * Note that kmalloc() lives in slab.c | |
6 | * | |
7 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
8 | * Swap reorganised 29.12.95, Stephen Tweedie | |
9 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
10 | * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999 | |
11 | * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999 | |
12 | * Zone balancing, Kanoj Sarcar, SGI, Jan 2000 | |
13 | * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002 | |
14 | * (lots of bits borrowed from Ingo Molnar & Andrew Morton) | |
15 | */ | |
16 | ||
1da177e4 LT |
17 | #include <linux/stddef.h> |
18 | #include <linux/mm.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/interrupt.h> | |
21 | #include <linux/pagemap.h> | |
22 | #include <linux/bootmem.h> | |
23 | #include <linux/compiler.h> | |
9f158333 | 24 | #include <linux/kernel.h> |
1da177e4 LT |
25 | #include <linux/module.h> |
26 | #include <linux/suspend.h> | |
27 | #include <linux/pagevec.h> | |
28 | #include <linux/blkdev.h> | |
29 | #include <linux/slab.h> | |
30 | #include <linux/notifier.h> | |
31 | #include <linux/topology.h> | |
32 | #include <linux/sysctl.h> | |
33 | #include <linux/cpu.h> | |
34 | #include <linux/cpuset.h> | |
bdc8cb98 | 35 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
36 | #include <linux/nodemask.h> |
37 | #include <linux/vmalloc.h> | |
4be38e35 | 38 | #include <linux/mempolicy.h> |
6811378e | 39 | #include <linux/stop_machine.h> |
c713216d MG |
40 | #include <linux/sort.h> |
41 | #include <linux/pfn.h> | |
1da177e4 LT |
42 | |
43 | #include <asm/tlbflush.h> | |
ac924c60 | 44 | #include <asm/div64.h> |
1da177e4 LT |
45 | #include "internal.h" |
46 | ||
47 | /* | |
48 | * MCD - HACK: Find somewhere to initialize this EARLY, or make this | |
49 | * initializer cleaner | |
50 | */ | |
c3d8c141 | 51 | nodemask_t node_online_map __read_mostly = { { [0] = 1UL } }; |
7223a93a | 52 | EXPORT_SYMBOL(node_online_map); |
c3d8c141 | 53 | nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL; |
7223a93a | 54 | EXPORT_SYMBOL(node_possible_map); |
6c231b7b | 55 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 56 | unsigned long totalreserve_pages __read_mostly; |
1da177e4 | 57 | long nr_swap_pages; |
8ad4b1fb | 58 | int percpu_pagelist_fraction; |
1da177e4 | 59 | |
d98c7a09 | 60 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 61 | |
1da177e4 LT |
62 | /* |
63 | * results with 256, 32 in the lowmem_reserve sysctl: | |
64 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
65 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
66 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
67 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
68 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
69 | * |
70 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
71 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 72 | */ |
2f1b6248 CL |
73 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
74 | 256, | |
fb0e7942 | 75 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 76 | 256, |
fb0e7942 | 77 | #endif |
e53ef38d | 78 | #ifdef CONFIG_HIGHMEM |
2f1b6248 | 79 | 32 |
e53ef38d | 80 | #endif |
2f1b6248 | 81 | }; |
1da177e4 LT |
82 | |
83 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 LT |
84 | |
85 | /* | |
86 | * Used by page_zone() to look up the address of the struct zone whose | |
87 | * id is encoded in the upper bits of page->flags | |
88 | */ | |
c3d8c141 | 89 | struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly; |
1da177e4 LT |
90 | EXPORT_SYMBOL(zone_table); |
91 | ||
2f1b6248 CL |
92 | static char *zone_names[MAX_NR_ZONES] = { |
93 | "DMA", | |
fb0e7942 | 94 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 95 | "DMA32", |
fb0e7942 | 96 | #endif |
2f1b6248 | 97 | "Normal", |
e53ef38d | 98 | #ifdef CONFIG_HIGHMEM |
2f1b6248 | 99 | "HighMem" |
e53ef38d | 100 | #endif |
2f1b6248 CL |
101 | }; |
102 | ||
1da177e4 LT |
103 | int min_free_kbytes = 1024; |
104 | ||
86356ab1 YG |
105 | unsigned long __meminitdata nr_kernel_pages; |
106 | unsigned long __meminitdata nr_all_pages; | |
0e0b864e | 107 | static unsigned long __initdata dma_reserve; |
1da177e4 | 108 | |
c713216d MG |
109 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
110 | /* | |
111 | * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct | |
112 | * ranges of memory (RAM) that may be registered with add_active_range(). | |
113 | * Ranges passed to add_active_range() will be merged if possible | |
114 | * so the number of times add_active_range() can be called is | |
115 | * related to the number of nodes and the number of holes | |
116 | */ | |
117 | #ifdef CONFIG_MAX_ACTIVE_REGIONS | |
118 | /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */ | |
119 | #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS | |
120 | #else | |
121 | #if MAX_NUMNODES >= 32 | |
122 | /* If there can be many nodes, allow up to 50 holes per node */ | |
123 | #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50) | |
124 | #else | |
125 | /* By default, allow up to 256 distinct regions */ | |
126 | #define MAX_ACTIVE_REGIONS 256 | |
127 | #endif | |
128 | #endif | |
129 | ||
130 | struct node_active_region __initdata early_node_map[MAX_ACTIVE_REGIONS]; | |
131 | int __initdata nr_nodemap_entries; | |
132 | unsigned long __initdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
133 | unsigned long __initdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
fb01439c MG |
134 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE |
135 | unsigned long __initdata node_boundary_start_pfn[MAX_NUMNODES]; | |
136 | unsigned long __initdata node_boundary_end_pfn[MAX_NUMNODES]; | |
137 | #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */ | |
c713216d MG |
138 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
139 | ||
13e7444b | 140 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 141 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 142 | { |
bdc8cb98 DH |
143 | int ret = 0; |
144 | unsigned seq; | |
145 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 146 | |
bdc8cb98 DH |
147 | do { |
148 | seq = zone_span_seqbegin(zone); | |
149 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
150 | ret = 1; | |
151 | else if (pfn < zone->zone_start_pfn) | |
152 | ret = 1; | |
153 | } while (zone_span_seqretry(zone, seq)); | |
154 | ||
155 | return ret; | |
c6a57e19 DH |
156 | } |
157 | ||
158 | static int page_is_consistent(struct zone *zone, struct page *page) | |
159 | { | |
1da177e4 LT |
160 | #ifdef CONFIG_HOLES_IN_ZONE |
161 | if (!pfn_valid(page_to_pfn(page))) | |
c6a57e19 | 162 | return 0; |
1da177e4 LT |
163 | #endif |
164 | if (zone != page_zone(page)) | |
c6a57e19 DH |
165 | return 0; |
166 | ||
167 | return 1; | |
168 | } | |
169 | /* | |
170 | * Temporary debugging check for pages not lying within a given zone. | |
171 | */ | |
172 | static int bad_range(struct zone *zone, struct page *page) | |
173 | { | |
174 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 175 | return 1; |
c6a57e19 DH |
176 | if (!page_is_consistent(zone, page)) |
177 | return 1; | |
178 | ||
1da177e4 LT |
179 | return 0; |
180 | } | |
13e7444b NP |
181 | #else |
182 | static inline int bad_range(struct zone *zone, struct page *page) | |
183 | { | |
184 | return 0; | |
185 | } | |
186 | #endif | |
187 | ||
224abf92 | 188 | static void bad_page(struct page *page) |
1da177e4 | 189 | { |
224abf92 | 190 | printk(KERN_EMERG "Bad page state in process '%s'\n" |
7365f3d1 HD |
191 | KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n" |
192 | KERN_EMERG "Trying to fix it up, but a reboot is needed\n" | |
193 | KERN_EMERG "Backtrace:\n", | |
224abf92 NP |
194 | current->comm, page, (int)(2*sizeof(unsigned long)), |
195 | (unsigned long)page->flags, page->mapping, | |
196 | page_mapcount(page), page_count(page)); | |
1da177e4 | 197 | dump_stack(); |
334795ec HD |
198 | page->flags &= ~(1 << PG_lru | |
199 | 1 << PG_private | | |
1da177e4 | 200 | 1 << PG_locked | |
1da177e4 LT |
201 | 1 << PG_active | |
202 | 1 << PG_dirty | | |
334795ec HD |
203 | 1 << PG_reclaim | |
204 | 1 << PG_slab | | |
1da177e4 | 205 | 1 << PG_swapcache | |
676165a8 NP |
206 | 1 << PG_writeback | |
207 | 1 << PG_buddy ); | |
1da177e4 LT |
208 | set_page_count(page, 0); |
209 | reset_page_mapcount(page); | |
210 | page->mapping = NULL; | |
9f158333 | 211 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
212 | } |
213 | ||
1da177e4 LT |
214 | /* |
215 | * Higher-order pages are called "compound pages". They are structured thusly: | |
216 | * | |
217 | * The first PAGE_SIZE page is called the "head page". | |
218 | * | |
219 | * The remaining PAGE_SIZE pages are called "tail pages". | |
220 | * | |
221 | * All pages have PG_compound set. All pages have their ->private pointing at | |
222 | * the head page (even the head page has this). | |
223 | * | |
41d78ba5 HD |
224 | * The first tail page's ->lru.next holds the address of the compound page's |
225 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
226 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 227 | */ |
d98c7a09 HD |
228 | |
229 | static void free_compound_page(struct page *page) | |
230 | { | |
231 | __free_pages_ok(page, (unsigned long)page[1].lru.prev); | |
232 | } | |
233 | ||
1da177e4 LT |
234 | static void prep_compound_page(struct page *page, unsigned long order) |
235 | { | |
236 | int i; | |
237 | int nr_pages = 1 << order; | |
238 | ||
d98c7a09 | 239 | page[1].lru.next = (void *)free_compound_page; /* set dtor */ |
41d78ba5 | 240 | page[1].lru.prev = (void *)order; |
1da177e4 LT |
241 | for (i = 0; i < nr_pages; i++) { |
242 | struct page *p = page + i; | |
243 | ||
5e9dace8 | 244 | __SetPageCompound(p); |
4c21e2f2 | 245 | set_page_private(p, (unsigned long)page); |
1da177e4 LT |
246 | } |
247 | } | |
248 | ||
249 | static void destroy_compound_page(struct page *page, unsigned long order) | |
250 | { | |
251 | int i; | |
252 | int nr_pages = 1 << order; | |
253 | ||
41d78ba5 | 254 | if (unlikely((unsigned long)page[1].lru.prev != order)) |
224abf92 | 255 | bad_page(page); |
1da177e4 LT |
256 | |
257 | for (i = 0; i < nr_pages; i++) { | |
258 | struct page *p = page + i; | |
259 | ||
224abf92 NP |
260 | if (unlikely(!PageCompound(p) | |
261 | (page_private(p) != (unsigned long)page))) | |
262 | bad_page(page); | |
5e9dace8 | 263 | __ClearPageCompound(p); |
1da177e4 LT |
264 | } |
265 | } | |
1da177e4 | 266 | |
17cf4406 NP |
267 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
268 | { | |
269 | int i; | |
270 | ||
725d704e | 271 | VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM); |
6626c5d5 AM |
272 | /* |
273 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
274 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
275 | */ | |
725d704e | 276 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
277 | for (i = 0; i < (1 << order); i++) |
278 | clear_highpage(page + i); | |
279 | } | |
280 | ||
1da177e4 LT |
281 | /* |
282 | * function for dealing with page's order in buddy system. | |
283 | * zone->lock is already acquired when we use these. | |
284 | * So, we don't need atomic page->flags operations here. | |
285 | */ | |
6aa3001b AM |
286 | static inline unsigned long page_order(struct page *page) |
287 | { | |
4c21e2f2 | 288 | return page_private(page); |
1da177e4 LT |
289 | } |
290 | ||
6aa3001b AM |
291 | static inline void set_page_order(struct page *page, int order) |
292 | { | |
4c21e2f2 | 293 | set_page_private(page, order); |
676165a8 | 294 | __SetPageBuddy(page); |
1da177e4 LT |
295 | } |
296 | ||
297 | static inline void rmv_page_order(struct page *page) | |
298 | { | |
676165a8 | 299 | __ClearPageBuddy(page); |
4c21e2f2 | 300 | set_page_private(page, 0); |
1da177e4 LT |
301 | } |
302 | ||
303 | /* | |
304 | * Locate the struct page for both the matching buddy in our | |
305 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
306 | * | |
307 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
308 | * the following equation: | |
309 | * B2 = B1 ^ (1 << O) | |
310 | * For example, if the starting buddy (buddy2) is #8 its order | |
311 | * 1 buddy is #10: | |
312 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
313 | * | |
314 | * 2) Any buddy B will have an order O+1 parent P which | |
315 | * satisfies the following equation: | |
316 | * P = B & ~(1 << O) | |
317 | * | |
d6e05edc | 318 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 LT |
319 | */ |
320 | static inline struct page * | |
321 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
322 | { | |
323 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
324 | ||
325 | return page + (buddy_idx - page_idx); | |
326 | } | |
327 | ||
328 | static inline unsigned long | |
329 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
330 | { | |
331 | return (page_idx & ~(1 << order)); | |
332 | } | |
333 | ||
334 | /* | |
335 | * This function checks whether a page is free && is the buddy | |
336 | * we can do coalesce a page and its buddy if | |
13e7444b | 337 | * (a) the buddy is not in a hole && |
676165a8 | 338 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
339 | * (c) a page and its buddy have the same order && |
340 | * (d) a page and its buddy are in the same zone. | |
676165a8 NP |
341 | * |
342 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
343 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
1da177e4 | 344 | * |
676165a8 | 345 | * For recording page's order, we use page_private(page). |
1da177e4 | 346 | */ |
cb2b95e1 AW |
347 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
348 | int order) | |
1da177e4 | 349 | { |
13e7444b | 350 | #ifdef CONFIG_HOLES_IN_ZONE |
cb2b95e1 | 351 | if (!pfn_valid(page_to_pfn(buddy))) |
13e7444b NP |
352 | return 0; |
353 | #endif | |
354 | ||
cb2b95e1 AW |
355 | if (page_zone_id(page) != page_zone_id(buddy)) |
356 | return 0; | |
357 | ||
358 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
359 | BUG_ON(page_count(buddy) != 0); | |
6aa3001b | 360 | return 1; |
676165a8 | 361 | } |
6aa3001b | 362 | return 0; |
1da177e4 LT |
363 | } |
364 | ||
365 | /* | |
366 | * Freeing function for a buddy system allocator. | |
367 | * | |
368 | * The concept of a buddy system is to maintain direct-mapped table | |
369 | * (containing bit values) for memory blocks of various "orders". | |
370 | * The bottom level table contains the map for the smallest allocatable | |
371 | * units of memory (here, pages), and each level above it describes | |
372 | * pairs of units from the levels below, hence, "buddies". | |
373 | * At a high level, all that happens here is marking the table entry | |
374 | * at the bottom level available, and propagating the changes upward | |
375 | * as necessary, plus some accounting needed to play nicely with other | |
376 | * parts of the VM system. | |
377 | * At each level, we keep a list of pages, which are heads of continuous | |
676165a8 | 378 | * free pages of length of (1 << order) and marked with PG_buddy. Page's |
4c21e2f2 | 379 | * order is recorded in page_private(page) field. |
1da177e4 LT |
380 | * So when we are allocating or freeing one, we can derive the state of the |
381 | * other. That is, if we allocate a small block, and both were | |
382 | * free, the remainder of the region must be split into blocks. | |
383 | * If a block is freed, and its buddy is also free, then this | |
384 | * triggers coalescing into a block of larger size. | |
385 | * | |
386 | * -- wli | |
387 | */ | |
388 | ||
48db57f8 | 389 | static inline void __free_one_page(struct page *page, |
1da177e4 LT |
390 | struct zone *zone, unsigned int order) |
391 | { | |
392 | unsigned long page_idx; | |
393 | int order_size = 1 << order; | |
394 | ||
224abf92 | 395 | if (unlikely(PageCompound(page))) |
1da177e4 LT |
396 | destroy_compound_page(page, order); |
397 | ||
398 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); | |
399 | ||
725d704e NP |
400 | VM_BUG_ON(page_idx & (order_size - 1)); |
401 | VM_BUG_ON(bad_range(zone, page)); | |
1da177e4 LT |
402 | |
403 | zone->free_pages += order_size; | |
404 | while (order < MAX_ORDER-1) { | |
405 | unsigned long combined_idx; | |
406 | struct free_area *area; | |
407 | struct page *buddy; | |
408 | ||
1da177e4 | 409 | buddy = __page_find_buddy(page, page_idx, order); |
cb2b95e1 | 410 | if (!page_is_buddy(page, buddy, order)) |
1da177e4 | 411 | break; /* Move the buddy up one level. */ |
13e7444b | 412 | |
1da177e4 LT |
413 | list_del(&buddy->lru); |
414 | area = zone->free_area + order; | |
415 | area->nr_free--; | |
416 | rmv_page_order(buddy); | |
13e7444b | 417 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
418 | page = page + (combined_idx - page_idx); |
419 | page_idx = combined_idx; | |
420 | order++; | |
421 | } | |
422 | set_page_order(page, order); | |
423 | list_add(&page->lru, &zone->free_area[order].free_list); | |
424 | zone->free_area[order].nr_free++; | |
425 | } | |
426 | ||
224abf92 | 427 | static inline int free_pages_check(struct page *page) |
1da177e4 | 428 | { |
92be2e33 NP |
429 | if (unlikely(page_mapcount(page) | |
430 | (page->mapping != NULL) | | |
431 | (page_count(page) != 0) | | |
1da177e4 LT |
432 | (page->flags & ( |
433 | 1 << PG_lru | | |
434 | 1 << PG_private | | |
435 | 1 << PG_locked | | |
436 | 1 << PG_active | | |
437 | 1 << PG_reclaim | | |
438 | 1 << PG_slab | | |
439 | 1 << PG_swapcache | | |
b5810039 | 440 | 1 << PG_writeback | |
676165a8 NP |
441 | 1 << PG_reserved | |
442 | 1 << PG_buddy )))) | |
224abf92 | 443 | bad_page(page); |
1da177e4 | 444 | if (PageDirty(page)) |
242e5468 | 445 | __ClearPageDirty(page); |
689bcebf HD |
446 | /* |
447 | * For now, we report if PG_reserved was found set, but do not | |
448 | * clear it, and do not free the page. But we shall soon need | |
449 | * to do more, for when the ZERO_PAGE count wraps negative. | |
450 | */ | |
451 | return PageReserved(page); | |
1da177e4 LT |
452 | } |
453 | ||
454 | /* | |
455 | * Frees a list of pages. | |
456 | * Assumes all pages on list are in same zone, and of same order. | |
207f36ee | 457 | * count is the number of pages to free. |
1da177e4 LT |
458 | * |
459 | * If the zone was previously in an "all pages pinned" state then look to | |
460 | * see if this freeing clears that state. | |
461 | * | |
462 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
463 | * pinned" detection logic. | |
464 | */ | |
48db57f8 NP |
465 | static void free_pages_bulk(struct zone *zone, int count, |
466 | struct list_head *list, int order) | |
1da177e4 | 467 | { |
c54ad30c | 468 | spin_lock(&zone->lock); |
1da177e4 LT |
469 | zone->all_unreclaimable = 0; |
470 | zone->pages_scanned = 0; | |
48db57f8 NP |
471 | while (count--) { |
472 | struct page *page; | |
473 | ||
725d704e | 474 | VM_BUG_ON(list_empty(list)); |
1da177e4 | 475 | page = list_entry(list->prev, struct page, lru); |
48db57f8 | 476 | /* have to delete it as __free_one_page list manipulates */ |
1da177e4 | 477 | list_del(&page->lru); |
48db57f8 | 478 | __free_one_page(page, zone, order); |
1da177e4 | 479 | } |
c54ad30c | 480 | spin_unlock(&zone->lock); |
1da177e4 LT |
481 | } |
482 | ||
48db57f8 | 483 | static void free_one_page(struct zone *zone, struct page *page, int order) |
1da177e4 | 484 | { |
006d22d9 CL |
485 | spin_lock(&zone->lock); |
486 | zone->all_unreclaimable = 0; | |
487 | zone->pages_scanned = 0; | |
488 | __free_one_page(page, zone ,order); | |
489 | spin_unlock(&zone->lock); | |
48db57f8 NP |
490 | } |
491 | ||
492 | static void __free_pages_ok(struct page *page, unsigned int order) | |
493 | { | |
494 | unsigned long flags; | |
1da177e4 | 495 | int i; |
689bcebf | 496 | int reserved = 0; |
1da177e4 | 497 | |
de5097c2 | 498 | if (!PageHighMem(page)) |
f9b8404c IM |
499 | debug_check_no_locks_freed(page_address(page), |
500 | PAGE_SIZE<<order); | |
1da177e4 | 501 | |
1da177e4 | 502 | for (i = 0 ; i < (1 << order) ; ++i) |
224abf92 | 503 | reserved += free_pages_check(page + i); |
689bcebf HD |
504 | if (reserved) |
505 | return; | |
506 | ||
dafb1367 | 507 | arch_free_page(page, order); |
48db57f8 | 508 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 509 | |
c54ad30c | 510 | local_irq_save(flags); |
f8891e5e | 511 | __count_vm_events(PGFREE, 1 << order); |
48db57f8 | 512 | free_one_page(page_zone(page), page, order); |
c54ad30c | 513 | local_irq_restore(flags); |
1da177e4 LT |
514 | } |
515 | ||
a226f6c8 DH |
516 | /* |
517 | * permit the bootmem allocator to evade page validation on high-order frees | |
518 | */ | |
519 | void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order) | |
520 | { | |
521 | if (order == 0) { | |
522 | __ClearPageReserved(page); | |
523 | set_page_count(page, 0); | |
7835e98b | 524 | set_page_refcounted(page); |
545b1ea9 | 525 | __free_page(page); |
a226f6c8 | 526 | } else { |
a226f6c8 DH |
527 | int loop; |
528 | ||
545b1ea9 | 529 | prefetchw(page); |
a226f6c8 DH |
530 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
531 | struct page *p = &page[loop]; | |
532 | ||
545b1ea9 NP |
533 | if (loop + 1 < BITS_PER_LONG) |
534 | prefetchw(p + 1); | |
a226f6c8 DH |
535 | __ClearPageReserved(p); |
536 | set_page_count(p, 0); | |
537 | } | |
538 | ||
7835e98b | 539 | set_page_refcounted(page); |
545b1ea9 | 540 | __free_pages(page, order); |
a226f6c8 DH |
541 | } |
542 | } | |
543 | ||
1da177e4 LT |
544 | |
545 | /* | |
546 | * The order of subdivision here is critical for the IO subsystem. | |
547 | * Please do not alter this order without good reasons and regression | |
548 | * testing. Specifically, as large blocks of memory are subdivided, | |
549 | * the order in which smaller blocks are delivered depends on the order | |
550 | * they're subdivided in this function. This is the primary factor | |
551 | * influencing the order in which pages are delivered to the IO | |
552 | * subsystem according to empirical testing, and this is also justified | |
553 | * by considering the behavior of a buddy system containing a single | |
554 | * large block of memory acted on by a series of small allocations. | |
555 | * This behavior is a critical factor in sglist merging's success. | |
556 | * | |
557 | * -- wli | |
558 | */ | |
085cc7d5 | 559 | static inline void expand(struct zone *zone, struct page *page, |
1da177e4 LT |
560 | int low, int high, struct free_area *area) |
561 | { | |
562 | unsigned long size = 1 << high; | |
563 | ||
564 | while (high > low) { | |
565 | area--; | |
566 | high--; | |
567 | size >>= 1; | |
725d704e | 568 | VM_BUG_ON(bad_range(zone, &page[size])); |
1da177e4 LT |
569 | list_add(&page[size].lru, &area->free_list); |
570 | area->nr_free++; | |
571 | set_page_order(&page[size], high); | |
572 | } | |
1da177e4 LT |
573 | } |
574 | ||
1da177e4 LT |
575 | /* |
576 | * This page is about to be returned from the page allocator | |
577 | */ | |
17cf4406 | 578 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) |
1da177e4 | 579 | { |
92be2e33 NP |
580 | if (unlikely(page_mapcount(page) | |
581 | (page->mapping != NULL) | | |
582 | (page_count(page) != 0) | | |
334795ec HD |
583 | (page->flags & ( |
584 | 1 << PG_lru | | |
1da177e4 LT |
585 | 1 << PG_private | |
586 | 1 << PG_locked | | |
1da177e4 LT |
587 | 1 << PG_active | |
588 | 1 << PG_dirty | | |
589 | 1 << PG_reclaim | | |
334795ec | 590 | 1 << PG_slab | |
1da177e4 | 591 | 1 << PG_swapcache | |
b5810039 | 592 | 1 << PG_writeback | |
676165a8 NP |
593 | 1 << PG_reserved | |
594 | 1 << PG_buddy )))) | |
224abf92 | 595 | bad_page(page); |
1da177e4 | 596 | |
689bcebf HD |
597 | /* |
598 | * For now, we report if PG_reserved was found set, but do not | |
599 | * clear it, and do not allocate the page: as a safety net. | |
600 | */ | |
601 | if (PageReserved(page)) | |
602 | return 1; | |
603 | ||
1da177e4 LT |
604 | page->flags &= ~(1 << PG_uptodate | 1 << PG_error | |
605 | 1 << PG_referenced | 1 << PG_arch_1 | | |
606 | 1 << PG_checked | 1 << PG_mappedtodisk); | |
4c21e2f2 | 607 | set_page_private(page, 0); |
7835e98b | 608 | set_page_refcounted(page); |
1da177e4 | 609 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
610 | |
611 | if (gfp_flags & __GFP_ZERO) | |
612 | prep_zero_page(page, order, gfp_flags); | |
613 | ||
614 | if (order && (gfp_flags & __GFP_COMP)) | |
615 | prep_compound_page(page, order); | |
616 | ||
689bcebf | 617 | return 0; |
1da177e4 LT |
618 | } |
619 | ||
620 | /* | |
621 | * Do the hard work of removing an element from the buddy allocator. | |
622 | * Call me with the zone->lock already held. | |
623 | */ | |
624 | static struct page *__rmqueue(struct zone *zone, unsigned int order) | |
625 | { | |
626 | struct free_area * area; | |
627 | unsigned int current_order; | |
628 | struct page *page; | |
629 | ||
630 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
631 | area = zone->free_area + current_order; | |
632 | if (list_empty(&area->free_list)) | |
633 | continue; | |
634 | ||
635 | page = list_entry(area->free_list.next, struct page, lru); | |
636 | list_del(&page->lru); | |
637 | rmv_page_order(page); | |
638 | area->nr_free--; | |
639 | zone->free_pages -= 1UL << order; | |
085cc7d5 NP |
640 | expand(zone, page, order, current_order, area); |
641 | return page; | |
1da177e4 LT |
642 | } |
643 | ||
644 | return NULL; | |
645 | } | |
646 | ||
647 | /* | |
648 | * Obtain a specified number of elements from the buddy allocator, all under | |
649 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
650 | * Returns the number of new pages which were placed at *list. | |
651 | */ | |
652 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
653 | unsigned long count, struct list_head *list) | |
654 | { | |
1da177e4 | 655 | int i; |
1da177e4 | 656 | |
c54ad30c | 657 | spin_lock(&zone->lock); |
1da177e4 | 658 | for (i = 0; i < count; ++i) { |
085cc7d5 NP |
659 | struct page *page = __rmqueue(zone, order); |
660 | if (unlikely(page == NULL)) | |
1da177e4 | 661 | break; |
1da177e4 LT |
662 | list_add_tail(&page->lru, list); |
663 | } | |
c54ad30c | 664 | spin_unlock(&zone->lock); |
085cc7d5 | 665 | return i; |
1da177e4 LT |
666 | } |
667 | ||
4ae7c039 | 668 | #ifdef CONFIG_NUMA |
8fce4d8e CL |
669 | /* |
670 | * Called from the slab reaper to drain pagesets on a particular node that | |
39bbcb8f | 671 | * belongs to the currently executing processor. |
879336c3 CL |
672 | * Note that this function must be called with the thread pinned to |
673 | * a single processor. | |
8fce4d8e CL |
674 | */ |
675 | void drain_node_pages(int nodeid) | |
4ae7c039 | 676 | { |
2f6726e5 CL |
677 | int i; |
678 | enum zone_type z; | |
4ae7c039 CL |
679 | unsigned long flags; |
680 | ||
8fce4d8e CL |
681 | for (z = 0; z < MAX_NR_ZONES; z++) { |
682 | struct zone *zone = NODE_DATA(nodeid)->node_zones + z; | |
4ae7c039 CL |
683 | struct per_cpu_pageset *pset; |
684 | ||
39bbcb8f CL |
685 | if (!populated_zone(zone)) |
686 | continue; | |
687 | ||
23316bc8 | 688 | pset = zone_pcp(zone, smp_processor_id()); |
4ae7c039 CL |
689 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
690 | struct per_cpu_pages *pcp; | |
691 | ||
692 | pcp = &pset->pcp[i]; | |
879336c3 CL |
693 | if (pcp->count) { |
694 | local_irq_save(flags); | |
695 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); | |
696 | pcp->count = 0; | |
697 | local_irq_restore(flags); | |
698 | } | |
4ae7c039 CL |
699 | } |
700 | } | |
4ae7c039 CL |
701 | } |
702 | #endif | |
703 | ||
1da177e4 LT |
704 | #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU) |
705 | static void __drain_pages(unsigned int cpu) | |
706 | { | |
c54ad30c | 707 | unsigned long flags; |
1da177e4 LT |
708 | struct zone *zone; |
709 | int i; | |
710 | ||
711 | for_each_zone(zone) { | |
712 | struct per_cpu_pageset *pset; | |
713 | ||
e7c8d5c9 | 714 | pset = zone_pcp(zone, cpu); |
1da177e4 LT |
715 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
716 | struct per_cpu_pages *pcp; | |
717 | ||
718 | pcp = &pset->pcp[i]; | |
c54ad30c | 719 | local_irq_save(flags); |
48db57f8 NP |
720 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); |
721 | pcp->count = 0; | |
c54ad30c | 722 | local_irq_restore(flags); |
1da177e4 LT |
723 | } |
724 | } | |
725 | } | |
726 | #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */ | |
727 | ||
728 | #ifdef CONFIG_PM | |
729 | ||
730 | void mark_free_pages(struct zone *zone) | |
731 | { | |
f623f0db RW |
732 | unsigned long pfn, max_zone_pfn; |
733 | unsigned long flags; | |
1da177e4 LT |
734 | int order; |
735 | struct list_head *curr; | |
736 | ||
737 | if (!zone->spanned_pages) | |
738 | return; | |
739 | ||
740 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db RW |
741 | |
742 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
743 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
744 | if (pfn_valid(pfn)) { | |
745 | struct page *page = pfn_to_page(pfn); | |
746 | ||
747 | if (!PageNosave(page)) | |
748 | ClearPageNosaveFree(page); | |
749 | } | |
1da177e4 LT |
750 | |
751 | for (order = MAX_ORDER - 1; order >= 0; --order) | |
752 | list_for_each(curr, &zone->free_area[order].free_list) { | |
f623f0db | 753 | unsigned long i; |
1da177e4 | 754 | |
f623f0db RW |
755 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
756 | for (i = 0; i < (1UL << order); i++) | |
757 | SetPageNosaveFree(pfn_to_page(pfn + i)); | |
758 | } | |
1da177e4 | 759 | |
1da177e4 LT |
760 | spin_unlock_irqrestore(&zone->lock, flags); |
761 | } | |
762 | ||
763 | /* | |
764 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
765 | */ | |
766 | void drain_local_pages(void) | |
767 | { | |
768 | unsigned long flags; | |
769 | ||
770 | local_irq_save(flags); | |
771 | __drain_pages(smp_processor_id()); | |
772 | local_irq_restore(flags); | |
773 | } | |
774 | #endif /* CONFIG_PM */ | |
775 | ||
1da177e4 LT |
776 | /* |
777 | * Free a 0-order page | |
778 | */ | |
1da177e4 LT |
779 | static void fastcall free_hot_cold_page(struct page *page, int cold) |
780 | { | |
781 | struct zone *zone = page_zone(page); | |
782 | struct per_cpu_pages *pcp; | |
783 | unsigned long flags; | |
784 | ||
1da177e4 LT |
785 | if (PageAnon(page)) |
786 | page->mapping = NULL; | |
224abf92 | 787 | if (free_pages_check(page)) |
689bcebf HD |
788 | return; |
789 | ||
dafb1367 | 790 | arch_free_page(page, 0); |
689bcebf HD |
791 | kernel_map_pages(page, 1, 0); |
792 | ||
e7c8d5c9 | 793 | pcp = &zone_pcp(zone, get_cpu())->pcp[cold]; |
1da177e4 | 794 | local_irq_save(flags); |
f8891e5e | 795 | __count_vm_event(PGFREE); |
1da177e4 LT |
796 | list_add(&page->lru, &pcp->list); |
797 | pcp->count++; | |
48db57f8 NP |
798 | if (pcp->count >= pcp->high) { |
799 | free_pages_bulk(zone, pcp->batch, &pcp->list, 0); | |
800 | pcp->count -= pcp->batch; | |
801 | } | |
1da177e4 LT |
802 | local_irq_restore(flags); |
803 | put_cpu(); | |
804 | } | |
805 | ||
806 | void fastcall free_hot_page(struct page *page) | |
807 | { | |
808 | free_hot_cold_page(page, 0); | |
809 | } | |
810 | ||
811 | void fastcall free_cold_page(struct page *page) | |
812 | { | |
813 | free_hot_cold_page(page, 1); | |
814 | } | |
815 | ||
8dfcc9ba NP |
816 | /* |
817 | * split_page takes a non-compound higher-order page, and splits it into | |
818 | * n (1<<order) sub-pages: page[0..n] | |
819 | * Each sub-page must be freed individually. | |
820 | * | |
821 | * Note: this is probably too low level an operation for use in drivers. | |
822 | * Please consult with lkml before using this in your driver. | |
823 | */ | |
824 | void split_page(struct page *page, unsigned int order) | |
825 | { | |
826 | int i; | |
827 | ||
725d704e NP |
828 | VM_BUG_ON(PageCompound(page)); |
829 | VM_BUG_ON(!page_count(page)); | |
7835e98b NP |
830 | for (i = 1; i < (1 << order); i++) |
831 | set_page_refcounted(page + i); | |
8dfcc9ba | 832 | } |
8dfcc9ba | 833 | |
1da177e4 LT |
834 | /* |
835 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
836 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
837 | * or two. | |
838 | */ | |
a74609fa NP |
839 | static struct page *buffered_rmqueue(struct zonelist *zonelist, |
840 | struct zone *zone, int order, gfp_t gfp_flags) | |
1da177e4 LT |
841 | { |
842 | unsigned long flags; | |
689bcebf | 843 | struct page *page; |
1da177e4 | 844 | int cold = !!(gfp_flags & __GFP_COLD); |
a74609fa | 845 | int cpu; |
1da177e4 | 846 | |
689bcebf | 847 | again: |
a74609fa | 848 | cpu = get_cpu(); |
48db57f8 | 849 | if (likely(order == 0)) { |
1da177e4 LT |
850 | struct per_cpu_pages *pcp; |
851 | ||
a74609fa | 852 | pcp = &zone_pcp(zone, cpu)->pcp[cold]; |
1da177e4 | 853 | local_irq_save(flags); |
a74609fa | 854 | if (!pcp->count) { |
1da177e4 LT |
855 | pcp->count += rmqueue_bulk(zone, 0, |
856 | pcp->batch, &pcp->list); | |
a74609fa NP |
857 | if (unlikely(!pcp->count)) |
858 | goto failed; | |
1da177e4 | 859 | } |
a74609fa NP |
860 | page = list_entry(pcp->list.next, struct page, lru); |
861 | list_del(&page->lru); | |
862 | pcp->count--; | |
7fb1d9fc | 863 | } else { |
1da177e4 LT |
864 | spin_lock_irqsave(&zone->lock, flags); |
865 | page = __rmqueue(zone, order); | |
a74609fa NP |
866 | spin_unlock(&zone->lock); |
867 | if (!page) | |
868 | goto failed; | |
1da177e4 LT |
869 | } |
870 | ||
f8891e5e | 871 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
ca889e6c | 872 | zone_statistics(zonelist, zone); |
a74609fa NP |
873 | local_irq_restore(flags); |
874 | put_cpu(); | |
1da177e4 | 875 | |
725d704e | 876 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 877 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 878 | goto again; |
1da177e4 | 879 | return page; |
a74609fa NP |
880 | |
881 | failed: | |
882 | local_irq_restore(flags); | |
883 | put_cpu(); | |
884 | return NULL; | |
1da177e4 LT |
885 | } |
886 | ||
7fb1d9fc | 887 | #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */ |
3148890b NP |
888 | #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */ |
889 | #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */ | |
890 | #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */ | |
891 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ | |
892 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
893 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 894 | |
1da177e4 LT |
895 | /* |
896 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
897 | * of the allocation. | |
898 | */ | |
899 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 900 | int classzone_idx, int alloc_flags) |
1da177e4 LT |
901 | { |
902 | /* free_pages my go negative - that's OK */ | |
e80ee884 NP |
903 | unsigned long min = mark; |
904 | long free_pages = z->free_pages - (1 << order) + 1; | |
1da177e4 LT |
905 | int o; |
906 | ||
7fb1d9fc | 907 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 908 | min -= min / 2; |
7fb1d9fc | 909 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
910 | min -= min / 4; |
911 | ||
912 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
913 | return 0; | |
914 | for (o = 0; o < order; o++) { | |
915 | /* At the next order, this order's pages become unavailable */ | |
916 | free_pages -= z->free_area[o].nr_free << o; | |
917 | ||
918 | /* Require fewer higher order pages to be free */ | |
919 | min >>= 1; | |
920 | ||
921 | if (free_pages <= min) | |
922 | return 0; | |
923 | } | |
924 | return 1; | |
925 | } | |
926 | ||
7fb1d9fc RS |
927 | /* |
928 | * get_page_from_freeliest goes through the zonelist trying to allocate | |
929 | * a page. | |
930 | */ | |
931 | static struct page * | |
932 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, | |
933 | struct zonelist *zonelist, int alloc_flags) | |
753ee728 | 934 | { |
7fb1d9fc RS |
935 | struct zone **z = zonelist->zones; |
936 | struct page *page = NULL; | |
937 | int classzone_idx = zone_idx(*z); | |
1192d526 | 938 | struct zone *zone; |
7fb1d9fc RS |
939 | |
940 | /* | |
941 | * Go through the zonelist once, looking for a zone with enough free. | |
942 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
943 | */ | |
944 | do { | |
1192d526 | 945 | zone = *z; |
08e0f6a9 | 946 | if (unlikely(NUMA_BUILD && (gfp_mask & __GFP_THISNODE) && |
1192d526 | 947 | zone->zone_pgdat != zonelist->zones[0]->zone_pgdat)) |
9b819d20 | 948 | break; |
7fb1d9fc | 949 | if ((alloc_flags & ALLOC_CPUSET) && |
1192d526 | 950 | !cpuset_zone_allowed(zone, gfp_mask)) |
7fb1d9fc RS |
951 | continue; |
952 | ||
953 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { | |
3148890b NP |
954 | unsigned long mark; |
955 | if (alloc_flags & ALLOC_WMARK_MIN) | |
1192d526 | 956 | mark = zone->pages_min; |
3148890b | 957 | else if (alloc_flags & ALLOC_WMARK_LOW) |
1192d526 | 958 | mark = zone->pages_low; |
3148890b | 959 | else |
1192d526 CL |
960 | mark = zone->pages_high; |
961 | if (!zone_watermark_ok(zone , order, mark, | |
7fb1d9fc | 962 | classzone_idx, alloc_flags)) |
9eeff239 | 963 | if (!zone_reclaim_mode || |
1192d526 | 964 | !zone_reclaim(zone, gfp_mask, order)) |
9eeff239 | 965 | continue; |
7fb1d9fc RS |
966 | } |
967 | ||
1192d526 | 968 | page = buffered_rmqueue(zonelist, zone, order, gfp_mask); |
7fb1d9fc | 969 | if (page) { |
7fb1d9fc RS |
970 | break; |
971 | } | |
972 | } while (*(++z) != NULL); | |
973 | return page; | |
753ee728 MH |
974 | } |
975 | ||
1da177e4 LT |
976 | /* |
977 | * This is the 'heart' of the zoned buddy allocator. | |
978 | */ | |
979 | struct page * fastcall | |
dd0fc66f | 980 | __alloc_pages(gfp_t gfp_mask, unsigned int order, |
1da177e4 LT |
981 | struct zonelist *zonelist) |
982 | { | |
260b2367 | 983 | const gfp_t wait = gfp_mask & __GFP_WAIT; |
7fb1d9fc | 984 | struct zone **z; |
1da177e4 LT |
985 | struct page *page; |
986 | struct reclaim_state reclaim_state; | |
987 | struct task_struct *p = current; | |
1da177e4 | 988 | int do_retry; |
7fb1d9fc | 989 | int alloc_flags; |
1da177e4 LT |
990 | int did_some_progress; |
991 | ||
992 | might_sleep_if(wait); | |
993 | ||
6b1de916 | 994 | restart: |
7fb1d9fc | 995 | z = zonelist->zones; /* the list of zones suitable for gfp_mask */ |
1da177e4 | 996 | |
7fb1d9fc | 997 | if (unlikely(*z == NULL)) { |
1da177e4 LT |
998 | /* Should this ever happen?? */ |
999 | return NULL; | |
1000 | } | |
6b1de916 | 1001 | |
7fb1d9fc | 1002 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1003 | zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET); |
7fb1d9fc RS |
1004 | if (page) |
1005 | goto got_pg; | |
1da177e4 | 1006 | |
6b1de916 | 1007 | do { |
43b0bc00 | 1008 | wakeup_kswapd(*z, order); |
6b1de916 | 1009 | } while (*(++z)); |
1da177e4 | 1010 | |
9bf2229f | 1011 | /* |
7fb1d9fc RS |
1012 | * OK, we're below the kswapd watermark and have kicked background |
1013 | * reclaim. Now things get more complex, so set up alloc_flags according | |
1014 | * to how we want to proceed. | |
1015 | * | |
1016 | * The caller may dip into page reserves a bit more if the caller | |
1017 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4eac915d PJ |
1018 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
1019 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
9bf2229f | 1020 | */ |
3148890b | 1021 | alloc_flags = ALLOC_WMARK_MIN; |
7fb1d9fc RS |
1022 | if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait) |
1023 | alloc_flags |= ALLOC_HARDER; | |
1024 | if (gfp_mask & __GFP_HIGH) | |
1025 | alloc_flags |= ALLOC_HIGH; | |
bdd804f4 PJ |
1026 | if (wait) |
1027 | alloc_flags |= ALLOC_CPUSET; | |
1da177e4 LT |
1028 | |
1029 | /* | |
1030 | * Go through the zonelist again. Let __GFP_HIGH and allocations | |
7fb1d9fc | 1031 | * coming from realtime tasks go deeper into reserves. |
1da177e4 LT |
1032 | * |
1033 | * This is the last chance, in general, before the goto nopage. | |
1034 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. | |
9bf2229f | 1035 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1da177e4 | 1036 | */ |
7fb1d9fc RS |
1037 | page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags); |
1038 | if (page) | |
1039 | goto got_pg; | |
1da177e4 LT |
1040 | |
1041 | /* This allocation should allow future memory freeing. */ | |
b84a35be NP |
1042 | |
1043 | if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE))) | |
1044 | && !in_interrupt()) { | |
1045 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { | |
885036d3 | 1046 | nofail_alloc: |
b84a35be | 1047 | /* go through the zonelist yet again, ignoring mins */ |
7fb1d9fc | 1048 | page = get_page_from_freelist(gfp_mask, order, |
47f3a867 | 1049 | zonelist, ALLOC_NO_WATERMARKS); |
7fb1d9fc RS |
1050 | if (page) |
1051 | goto got_pg; | |
885036d3 KK |
1052 | if (gfp_mask & __GFP_NOFAIL) { |
1053 | blk_congestion_wait(WRITE, HZ/50); | |
1054 | goto nofail_alloc; | |
1055 | } | |
1da177e4 LT |
1056 | } |
1057 | goto nopage; | |
1058 | } | |
1059 | ||
1060 | /* Atomic allocations - we can't balance anything */ | |
1061 | if (!wait) | |
1062 | goto nopage; | |
1063 | ||
1064 | rebalance: | |
1065 | cond_resched(); | |
1066 | ||
1067 | /* We now go into synchronous reclaim */ | |
3e0d98b9 | 1068 | cpuset_memory_pressure_bump(); |
1da177e4 LT |
1069 | p->flags |= PF_MEMALLOC; |
1070 | reclaim_state.reclaimed_slab = 0; | |
1071 | p->reclaim_state = &reclaim_state; | |
1072 | ||
7fb1d9fc | 1073 | did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask); |
1da177e4 LT |
1074 | |
1075 | p->reclaim_state = NULL; | |
1076 | p->flags &= ~PF_MEMALLOC; | |
1077 | ||
1078 | cond_resched(); | |
1079 | ||
1080 | if (likely(did_some_progress)) { | |
7fb1d9fc RS |
1081 | page = get_page_from_freelist(gfp_mask, order, |
1082 | zonelist, alloc_flags); | |
1083 | if (page) | |
1084 | goto got_pg; | |
1da177e4 LT |
1085 | } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { |
1086 | /* | |
1087 | * Go through the zonelist yet one more time, keep | |
1088 | * very high watermark here, this is only to catch | |
1089 | * a parallel oom killing, we must fail if we're still | |
1090 | * under heavy pressure. | |
1091 | */ | |
7fb1d9fc | 1092 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1093 | zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET); |
7fb1d9fc RS |
1094 | if (page) |
1095 | goto got_pg; | |
1da177e4 | 1096 | |
9b0f8b04 | 1097 | out_of_memory(zonelist, gfp_mask, order); |
1da177e4 LT |
1098 | goto restart; |
1099 | } | |
1100 | ||
1101 | /* | |
1102 | * Don't let big-order allocations loop unless the caller explicitly | |
1103 | * requests that. Wait for some write requests to complete then retry. | |
1104 | * | |
1105 | * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order | |
1106 | * <= 3, but that may not be true in other implementations. | |
1107 | */ | |
1108 | do_retry = 0; | |
1109 | if (!(gfp_mask & __GFP_NORETRY)) { | |
1110 | if ((order <= 3) || (gfp_mask & __GFP_REPEAT)) | |
1111 | do_retry = 1; | |
1112 | if (gfp_mask & __GFP_NOFAIL) | |
1113 | do_retry = 1; | |
1114 | } | |
1115 | if (do_retry) { | |
1116 | blk_congestion_wait(WRITE, HZ/50); | |
1117 | goto rebalance; | |
1118 | } | |
1119 | ||
1120 | nopage: | |
1121 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1122 | printk(KERN_WARNING "%s: page allocation failure." | |
1123 | " order:%d, mode:0x%x\n", | |
1124 | p->comm, order, gfp_mask); | |
1125 | dump_stack(); | |
578c2fd6 | 1126 | show_mem(); |
1da177e4 | 1127 | } |
1da177e4 | 1128 | got_pg: |
1da177e4 LT |
1129 | return page; |
1130 | } | |
1131 | ||
1132 | EXPORT_SYMBOL(__alloc_pages); | |
1133 | ||
1134 | /* | |
1135 | * Common helper functions. | |
1136 | */ | |
dd0fc66f | 1137 | fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 LT |
1138 | { |
1139 | struct page * page; | |
1140 | page = alloc_pages(gfp_mask, order); | |
1141 | if (!page) | |
1142 | return 0; | |
1143 | return (unsigned long) page_address(page); | |
1144 | } | |
1145 | ||
1146 | EXPORT_SYMBOL(__get_free_pages); | |
1147 | ||
dd0fc66f | 1148 | fastcall unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 LT |
1149 | { |
1150 | struct page * page; | |
1151 | ||
1152 | /* | |
1153 | * get_zeroed_page() returns a 32-bit address, which cannot represent | |
1154 | * a highmem page | |
1155 | */ | |
725d704e | 1156 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
1da177e4 LT |
1157 | |
1158 | page = alloc_pages(gfp_mask | __GFP_ZERO, 0); | |
1159 | if (page) | |
1160 | return (unsigned long) page_address(page); | |
1161 | return 0; | |
1162 | } | |
1163 | ||
1164 | EXPORT_SYMBOL(get_zeroed_page); | |
1165 | ||
1166 | void __pagevec_free(struct pagevec *pvec) | |
1167 | { | |
1168 | int i = pagevec_count(pvec); | |
1169 | ||
1170 | while (--i >= 0) | |
1171 | free_hot_cold_page(pvec->pages[i], pvec->cold); | |
1172 | } | |
1173 | ||
1174 | fastcall void __free_pages(struct page *page, unsigned int order) | |
1175 | { | |
b5810039 | 1176 | if (put_page_testzero(page)) { |
1da177e4 LT |
1177 | if (order == 0) |
1178 | free_hot_page(page); | |
1179 | else | |
1180 | __free_pages_ok(page, order); | |
1181 | } | |
1182 | } | |
1183 | ||
1184 | EXPORT_SYMBOL(__free_pages); | |
1185 | ||
1186 | fastcall void free_pages(unsigned long addr, unsigned int order) | |
1187 | { | |
1188 | if (addr != 0) { | |
725d704e | 1189 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
1190 | __free_pages(virt_to_page((void *)addr), order); |
1191 | } | |
1192 | } | |
1193 | ||
1194 | EXPORT_SYMBOL(free_pages); | |
1195 | ||
1196 | /* | |
1197 | * Total amount of free (allocatable) RAM: | |
1198 | */ | |
1199 | unsigned int nr_free_pages(void) | |
1200 | { | |
1201 | unsigned int sum = 0; | |
1202 | struct zone *zone; | |
1203 | ||
1204 | for_each_zone(zone) | |
1205 | sum += zone->free_pages; | |
1206 | ||
1207 | return sum; | |
1208 | } | |
1209 | ||
1210 | EXPORT_SYMBOL(nr_free_pages); | |
1211 | ||
1212 | #ifdef CONFIG_NUMA | |
1213 | unsigned int nr_free_pages_pgdat(pg_data_t *pgdat) | |
1214 | { | |
2f6726e5 CL |
1215 | unsigned int sum = 0; |
1216 | enum zone_type i; | |
1da177e4 LT |
1217 | |
1218 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1219 | sum += pgdat->node_zones[i].free_pages; | |
1220 | ||
1221 | return sum; | |
1222 | } | |
1223 | #endif | |
1224 | ||
1225 | static unsigned int nr_free_zone_pages(int offset) | |
1226 | { | |
e310fd43 MB |
1227 | /* Just pick one node, since fallback list is circular */ |
1228 | pg_data_t *pgdat = NODE_DATA(numa_node_id()); | |
1da177e4 LT |
1229 | unsigned int sum = 0; |
1230 | ||
e310fd43 MB |
1231 | struct zonelist *zonelist = pgdat->node_zonelists + offset; |
1232 | struct zone **zonep = zonelist->zones; | |
1233 | struct zone *zone; | |
1da177e4 | 1234 | |
e310fd43 MB |
1235 | for (zone = *zonep++; zone; zone = *zonep++) { |
1236 | unsigned long size = zone->present_pages; | |
1237 | unsigned long high = zone->pages_high; | |
1238 | if (size > high) | |
1239 | sum += size - high; | |
1da177e4 LT |
1240 | } |
1241 | ||
1242 | return sum; | |
1243 | } | |
1244 | ||
1245 | /* | |
1246 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
1247 | */ | |
1248 | unsigned int nr_free_buffer_pages(void) | |
1249 | { | |
af4ca457 | 1250 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 LT |
1251 | } |
1252 | ||
1253 | /* | |
1254 | * Amount of free RAM allocatable within all zones | |
1255 | */ | |
1256 | unsigned int nr_free_pagecache_pages(void) | |
1257 | { | |
af4ca457 | 1258 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER)); |
1da177e4 | 1259 | } |
08e0f6a9 CL |
1260 | |
1261 | static inline void show_node(struct zone *zone) | |
1da177e4 | 1262 | { |
08e0f6a9 CL |
1263 | if (NUMA_BUILD) |
1264 | printk("Node %ld ", zone_to_nid(zone)); | |
1da177e4 | 1265 | } |
1da177e4 | 1266 | |
1da177e4 LT |
1267 | void si_meminfo(struct sysinfo *val) |
1268 | { | |
1269 | val->totalram = totalram_pages; | |
1270 | val->sharedram = 0; | |
1271 | val->freeram = nr_free_pages(); | |
1272 | val->bufferram = nr_blockdev_pages(); | |
1da177e4 LT |
1273 | val->totalhigh = totalhigh_pages; |
1274 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
1275 | val->mem_unit = PAGE_SIZE; |
1276 | } | |
1277 | ||
1278 | EXPORT_SYMBOL(si_meminfo); | |
1279 | ||
1280 | #ifdef CONFIG_NUMA | |
1281 | void si_meminfo_node(struct sysinfo *val, int nid) | |
1282 | { | |
1283 | pg_data_t *pgdat = NODE_DATA(nid); | |
1284 | ||
1285 | val->totalram = pgdat->node_present_pages; | |
1286 | val->freeram = nr_free_pages_pgdat(pgdat); | |
98d2b0eb | 1287 | #ifdef CONFIG_HIGHMEM |
1da177e4 LT |
1288 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
1289 | val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages; | |
98d2b0eb CL |
1290 | #else |
1291 | val->totalhigh = 0; | |
1292 | val->freehigh = 0; | |
1293 | #endif | |
1da177e4 LT |
1294 | val->mem_unit = PAGE_SIZE; |
1295 | } | |
1296 | #endif | |
1297 | ||
1298 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
1299 | ||
1300 | /* | |
1301 | * Show free area list (used inside shift_scroll-lock stuff) | |
1302 | * We also calculate the percentage fragmentation. We do this by counting the | |
1303 | * memory on each free list with the exception of the first item on the list. | |
1304 | */ | |
1305 | void show_free_areas(void) | |
1306 | { | |
c7241913 | 1307 | int cpu; |
1da177e4 LT |
1308 | unsigned long active; |
1309 | unsigned long inactive; | |
1310 | unsigned long free; | |
1311 | struct zone *zone; | |
1312 | ||
1313 | for_each_zone(zone) { | |
c7241913 | 1314 | if (!populated_zone(zone)) |
1da177e4 | 1315 | continue; |
c7241913 JS |
1316 | |
1317 | show_node(zone); | |
1318 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 1319 | |
6b482c67 | 1320 | for_each_online_cpu(cpu) { |
1da177e4 LT |
1321 | struct per_cpu_pageset *pageset; |
1322 | ||
e7c8d5c9 | 1323 | pageset = zone_pcp(zone, cpu); |
1da177e4 | 1324 | |
c7241913 JS |
1325 | printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d " |
1326 | "Cold: hi:%5d, btch:%4d usd:%4d\n", | |
1327 | cpu, pageset->pcp[0].high, | |
1328 | pageset->pcp[0].batch, pageset->pcp[0].count, | |
1329 | pageset->pcp[1].high, pageset->pcp[1].batch, | |
1330 | pageset->pcp[1].count); | |
1da177e4 LT |
1331 | } |
1332 | } | |
1333 | ||
1da177e4 LT |
1334 | get_zone_counts(&active, &inactive, &free); |
1335 | ||
1da177e4 LT |
1336 | printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu " |
1337 | "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n", | |
1338 | active, | |
1339 | inactive, | |
b1e7a8fd | 1340 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 1341 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 1342 | global_page_state(NR_UNSTABLE_NFS), |
1da177e4 | 1343 | nr_free_pages(), |
972d1a7b CL |
1344 | global_page_state(NR_SLAB_RECLAIMABLE) + |
1345 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 1346 | global_page_state(NR_FILE_MAPPED), |
df849a15 | 1347 | global_page_state(NR_PAGETABLE)); |
1da177e4 LT |
1348 | |
1349 | for_each_zone(zone) { | |
1350 | int i; | |
1351 | ||
c7241913 JS |
1352 | if (!populated_zone(zone)) |
1353 | continue; | |
1354 | ||
1da177e4 LT |
1355 | show_node(zone); |
1356 | printk("%s" | |
1357 | " free:%lukB" | |
1358 | " min:%lukB" | |
1359 | " low:%lukB" | |
1360 | " high:%lukB" | |
1361 | " active:%lukB" | |
1362 | " inactive:%lukB" | |
1363 | " present:%lukB" | |
1364 | " pages_scanned:%lu" | |
1365 | " all_unreclaimable? %s" | |
1366 | "\n", | |
1367 | zone->name, | |
1368 | K(zone->free_pages), | |
1369 | K(zone->pages_min), | |
1370 | K(zone->pages_low), | |
1371 | K(zone->pages_high), | |
1372 | K(zone->nr_active), | |
1373 | K(zone->nr_inactive), | |
1374 | K(zone->present_pages), | |
1375 | zone->pages_scanned, | |
1376 | (zone->all_unreclaimable ? "yes" : "no") | |
1377 | ); | |
1378 | printk("lowmem_reserve[]:"); | |
1379 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1380 | printk(" %lu", zone->lowmem_reserve[i]); | |
1381 | printk("\n"); | |
1382 | } | |
1383 | ||
1384 | for_each_zone(zone) { | |
8f9de51a | 1385 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
1da177e4 | 1386 | |
c7241913 JS |
1387 | if (!populated_zone(zone)) |
1388 | continue; | |
1389 | ||
1da177e4 LT |
1390 | show_node(zone); |
1391 | printk("%s: ", zone->name); | |
1da177e4 LT |
1392 | |
1393 | spin_lock_irqsave(&zone->lock, flags); | |
1394 | for (order = 0; order < MAX_ORDER; order++) { | |
8f9de51a KK |
1395 | nr[order] = zone->free_area[order].nr_free; |
1396 | total += nr[order] << order; | |
1da177e4 LT |
1397 | } |
1398 | spin_unlock_irqrestore(&zone->lock, flags); | |
8f9de51a KK |
1399 | for (order = 0; order < MAX_ORDER; order++) |
1400 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1da177e4 LT |
1401 | printk("= %lukB\n", K(total)); |
1402 | } | |
1403 | ||
1404 | show_swap_cache_info(); | |
1405 | } | |
1406 | ||
1407 | /* | |
1408 | * Builds allocation fallback zone lists. | |
1a93205b CL |
1409 | * |
1410 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 1411 | */ |
86356ab1 | 1412 | static int __meminit build_zonelists_node(pg_data_t *pgdat, |
2f6726e5 | 1413 | struct zonelist *zonelist, int nr_zones, enum zone_type zone_type) |
1da177e4 | 1414 | { |
1a93205b CL |
1415 | struct zone *zone; |
1416 | ||
98d2b0eb | 1417 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 1418 | zone_type++; |
02a68a5e CL |
1419 | |
1420 | do { | |
2f6726e5 | 1421 | zone_type--; |
070f8032 | 1422 | zone = pgdat->node_zones + zone_type; |
1a93205b | 1423 | if (populated_zone(zone)) { |
070f8032 CL |
1424 | zonelist->zones[nr_zones++] = zone; |
1425 | check_highest_zone(zone_type); | |
1da177e4 | 1426 | } |
02a68a5e | 1427 | |
2f6726e5 | 1428 | } while (zone_type); |
070f8032 | 1429 | return nr_zones; |
1da177e4 LT |
1430 | } |
1431 | ||
1432 | #ifdef CONFIG_NUMA | |
1433 | #define MAX_NODE_LOAD (num_online_nodes()) | |
86356ab1 | 1434 | static int __meminitdata node_load[MAX_NUMNODES]; |
1da177e4 | 1435 | /** |
4dc3b16b | 1436 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
1437 | * @node: node whose fallback list we're appending |
1438 | * @used_node_mask: nodemask_t of already used nodes | |
1439 | * | |
1440 | * We use a number of factors to determine which is the next node that should | |
1441 | * appear on a given node's fallback list. The node should not have appeared | |
1442 | * already in @node's fallback list, and it should be the next closest node | |
1443 | * according to the distance array (which contains arbitrary distance values | |
1444 | * from each node to each node in the system), and should also prefer nodes | |
1445 | * with no CPUs, since presumably they'll have very little allocation pressure | |
1446 | * on them otherwise. | |
1447 | * It returns -1 if no node is found. | |
1448 | */ | |
86356ab1 | 1449 | static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 1450 | { |
4cf808eb | 1451 | int n, val; |
1da177e4 LT |
1452 | int min_val = INT_MAX; |
1453 | int best_node = -1; | |
1454 | ||
4cf808eb LT |
1455 | /* Use the local node if we haven't already */ |
1456 | if (!node_isset(node, *used_node_mask)) { | |
1457 | node_set(node, *used_node_mask); | |
1458 | return node; | |
1459 | } | |
1da177e4 | 1460 | |
4cf808eb LT |
1461 | for_each_online_node(n) { |
1462 | cpumask_t tmp; | |
1da177e4 LT |
1463 | |
1464 | /* Don't want a node to appear more than once */ | |
1465 | if (node_isset(n, *used_node_mask)) | |
1466 | continue; | |
1467 | ||
1da177e4 LT |
1468 | /* Use the distance array to find the distance */ |
1469 | val = node_distance(node, n); | |
1470 | ||
4cf808eb LT |
1471 | /* Penalize nodes under us ("prefer the next node") */ |
1472 | val += (n < node); | |
1473 | ||
1da177e4 LT |
1474 | /* Give preference to headless and unused nodes */ |
1475 | tmp = node_to_cpumask(n); | |
1476 | if (!cpus_empty(tmp)) | |
1477 | val += PENALTY_FOR_NODE_WITH_CPUS; | |
1478 | ||
1479 | /* Slight preference for less loaded node */ | |
1480 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
1481 | val += node_load[n]; | |
1482 | ||
1483 | if (val < min_val) { | |
1484 | min_val = val; | |
1485 | best_node = n; | |
1486 | } | |
1487 | } | |
1488 | ||
1489 | if (best_node >= 0) | |
1490 | node_set(best_node, *used_node_mask); | |
1491 | ||
1492 | return best_node; | |
1493 | } | |
1494 | ||
86356ab1 | 1495 | static void __meminit build_zonelists(pg_data_t *pgdat) |
1da177e4 | 1496 | { |
19655d34 CL |
1497 | int j, node, local_node; |
1498 | enum zone_type i; | |
1da177e4 LT |
1499 | int prev_node, load; |
1500 | struct zonelist *zonelist; | |
1501 | nodemask_t used_mask; | |
1502 | ||
1503 | /* initialize zonelists */ | |
19655d34 | 1504 | for (i = 0; i < MAX_NR_ZONES; i++) { |
1da177e4 LT |
1505 | zonelist = pgdat->node_zonelists + i; |
1506 | zonelist->zones[0] = NULL; | |
1507 | } | |
1508 | ||
1509 | /* NUMA-aware ordering of nodes */ | |
1510 | local_node = pgdat->node_id; | |
1511 | load = num_online_nodes(); | |
1512 | prev_node = local_node; | |
1513 | nodes_clear(used_mask); | |
1514 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { | |
9eeff239 CL |
1515 | int distance = node_distance(local_node, node); |
1516 | ||
1517 | /* | |
1518 | * If another node is sufficiently far away then it is better | |
1519 | * to reclaim pages in a zone before going off node. | |
1520 | */ | |
1521 | if (distance > RECLAIM_DISTANCE) | |
1522 | zone_reclaim_mode = 1; | |
1523 | ||
1da177e4 LT |
1524 | /* |
1525 | * We don't want to pressure a particular node. | |
1526 | * So adding penalty to the first node in same | |
1527 | * distance group to make it round-robin. | |
1528 | */ | |
9eeff239 CL |
1529 | |
1530 | if (distance != node_distance(local_node, prev_node)) | |
1da177e4 LT |
1531 | node_load[node] += load; |
1532 | prev_node = node; | |
1533 | load--; | |
19655d34 | 1534 | for (i = 0; i < MAX_NR_ZONES; i++) { |
1da177e4 LT |
1535 | zonelist = pgdat->node_zonelists + i; |
1536 | for (j = 0; zonelist->zones[j] != NULL; j++); | |
1537 | ||
19655d34 | 1538 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
1539 | zonelist->zones[j] = NULL; |
1540 | } | |
1541 | } | |
1542 | } | |
1543 | ||
1544 | #else /* CONFIG_NUMA */ | |
1545 | ||
86356ab1 | 1546 | static void __meminit build_zonelists(pg_data_t *pgdat) |
1da177e4 | 1547 | { |
19655d34 CL |
1548 | int node, local_node; |
1549 | enum zone_type i,j; | |
1da177e4 LT |
1550 | |
1551 | local_node = pgdat->node_id; | |
19655d34 | 1552 | for (i = 0; i < MAX_NR_ZONES; i++) { |
1da177e4 LT |
1553 | struct zonelist *zonelist; |
1554 | ||
1555 | zonelist = pgdat->node_zonelists + i; | |
1556 | ||
19655d34 | 1557 | j = build_zonelists_node(pgdat, zonelist, 0, i); |
1da177e4 LT |
1558 | /* |
1559 | * Now we build the zonelist so that it contains the zones | |
1560 | * of all the other nodes. | |
1561 | * We don't want to pressure a particular node, so when | |
1562 | * building the zones for node N, we make sure that the | |
1563 | * zones coming right after the local ones are those from | |
1564 | * node N+1 (modulo N) | |
1565 | */ | |
1566 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
1567 | if (!node_online(node)) | |
1568 | continue; | |
19655d34 | 1569 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
1570 | } |
1571 | for (node = 0; node < local_node; node++) { | |
1572 | if (!node_online(node)) | |
1573 | continue; | |
19655d34 | 1574 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
1575 | } |
1576 | ||
1577 | zonelist->zones[j] = NULL; | |
1578 | } | |
1579 | } | |
1580 | ||
1581 | #endif /* CONFIG_NUMA */ | |
1582 | ||
6811378e YG |
1583 | /* return values int ....just for stop_machine_run() */ |
1584 | static int __meminit __build_all_zonelists(void *dummy) | |
1da177e4 | 1585 | { |
6811378e YG |
1586 | int nid; |
1587 | for_each_online_node(nid) | |
1588 | build_zonelists(NODE_DATA(nid)); | |
1589 | return 0; | |
1590 | } | |
1591 | ||
1592 | void __meminit build_all_zonelists(void) | |
1593 | { | |
1594 | if (system_state == SYSTEM_BOOTING) { | |
423b41d7 | 1595 | __build_all_zonelists(NULL); |
6811378e YG |
1596 | cpuset_init_current_mems_allowed(); |
1597 | } else { | |
1598 | /* we have to stop all cpus to guaranntee there is no user | |
1599 | of zonelist */ | |
1600 | stop_machine_run(__build_all_zonelists, NULL, NR_CPUS); | |
1601 | /* cpuset refresh routine should be here */ | |
1602 | } | |
bd1e22b8 AM |
1603 | vm_total_pages = nr_free_pagecache_pages(); |
1604 | printk("Built %i zonelists. Total pages: %ld\n", | |
1605 | num_online_nodes(), vm_total_pages); | |
1da177e4 LT |
1606 | } |
1607 | ||
1608 | /* | |
1609 | * Helper functions to size the waitqueue hash table. | |
1610 | * Essentially these want to choose hash table sizes sufficiently | |
1611 | * large so that collisions trying to wait on pages are rare. | |
1612 | * But in fact, the number of active page waitqueues on typical | |
1613 | * systems is ridiculously low, less than 200. So this is even | |
1614 | * conservative, even though it seems large. | |
1615 | * | |
1616 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
1617 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
1618 | */ | |
1619 | #define PAGES_PER_WAITQUEUE 256 | |
1620 | ||
cca448fe | 1621 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 1622 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
1623 | { |
1624 | unsigned long size = 1; | |
1625 | ||
1626 | pages /= PAGES_PER_WAITQUEUE; | |
1627 | ||
1628 | while (size < pages) | |
1629 | size <<= 1; | |
1630 | ||
1631 | /* | |
1632 | * Once we have dozens or even hundreds of threads sleeping | |
1633 | * on IO we've got bigger problems than wait queue collision. | |
1634 | * Limit the size of the wait table to a reasonable size. | |
1635 | */ | |
1636 | size = min(size, 4096UL); | |
1637 | ||
1638 | return max(size, 4UL); | |
1639 | } | |
cca448fe YG |
1640 | #else |
1641 | /* | |
1642 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
1643 | * a suitable size for its wait_table. So we use the maximum size now. | |
1644 | * | |
1645 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
1646 | * | |
1647 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
1648 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
1649 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
1650 | * | |
1651 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
1652 | * or more by the traditional way. (See above). It equals: | |
1653 | * | |
1654 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
1655 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
1656 | * powerpc (64K page size) : = (32G +16M)byte. | |
1657 | */ | |
1658 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
1659 | { | |
1660 | return 4096UL; | |
1661 | } | |
1662 | #endif | |
1da177e4 LT |
1663 | |
1664 | /* | |
1665 | * This is an integer logarithm so that shifts can be used later | |
1666 | * to extract the more random high bits from the multiplicative | |
1667 | * hash function before the remainder is taken. | |
1668 | */ | |
1669 | static inline unsigned long wait_table_bits(unsigned long size) | |
1670 | { | |
1671 | return ffz(~size); | |
1672 | } | |
1673 | ||
1674 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
1675 | ||
1da177e4 LT |
1676 | /* |
1677 | * Initially all pages are reserved - free ones are freed | |
1678 | * up by free_all_bootmem() once the early boot process is | |
1679 | * done. Non-atomic initialization, single-pass. | |
1680 | */ | |
c09b4240 | 1681 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
1da177e4 LT |
1682 | unsigned long start_pfn) |
1683 | { | |
1da177e4 | 1684 | struct page *page; |
29751f69 AW |
1685 | unsigned long end_pfn = start_pfn + size; |
1686 | unsigned long pfn; | |
1da177e4 | 1687 | |
cbe8dd4a | 1688 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
d41dee36 AW |
1689 | if (!early_pfn_valid(pfn)) |
1690 | continue; | |
1691 | page = pfn_to_page(pfn); | |
1692 | set_page_links(page, zone, nid, pfn); | |
7835e98b | 1693 | init_page_count(page); |
1da177e4 LT |
1694 | reset_page_mapcount(page); |
1695 | SetPageReserved(page); | |
1696 | INIT_LIST_HEAD(&page->lru); | |
1697 | #ifdef WANT_PAGE_VIRTUAL | |
1698 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1699 | if (!is_highmem_idx(zone)) | |
3212c6be | 1700 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 1701 | #endif |
1da177e4 LT |
1702 | } |
1703 | } | |
1704 | ||
1705 | void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone, | |
1706 | unsigned long size) | |
1707 | { | |
1708 | int order; | |
1709 | for (order = 0; order < MAX_ORDER ; order++) { | |
1710 | INIT_LIST_HEAD(&zone->free_area[order].free_list); | |
1711 | zone->free_area[order].nr_free = 0; | |
1712 | } | |
1713 | } | |
1714 | ||
d41dee36 | 1715 | #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr) |
2f1b6248 CL |
1716 | void zonetable_add(struct zone *zone, int nid, enum zone_type zid, |
1717 | unsigned long pfn, unsigned long size) | |
d41dee36 AW |
1718 | { |
1719 | unsigned long snum = pfn_to_section_nr(pfn); | |
1720 | unsigned long end = pfn_to_section_nr(pfn + size); | |
1721 | ||
1722 | if (FLAGS_HAS_NODE) | |
1723 | zone_table[ZONETABLE_INDEX(nid, zid)] = zone; | |
1724 | else | |
1725 | for (; snum <= end; snum++) | |
1726 | zone_table[ZONETABLE_INDEX(snum, zid)] = zone; | |
1727 | } | |
1728 | ||
1da177e4 LT |
1729 | #ifndef __HAVE_ARCH_MEMMAP_INIT |
1730 | #define memmap_init(size, nid, zone, start_pfn) \ | |
1731 | memmap_init_zone((size), (nid), (zone), (start_pfn)) | |
1732 | #endif | |
1733 | ||
6292d9aa | 1734 | static int __cpuinit zone_batchsize(struct zone *zone) |
e7c8d5c9 CL |
1735 | { |
1736 | int batch; | |
1737 | ||
1738 | /* | |
1739 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 1740 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
1741 | * |
1742 | * OK, so we don't know how big the cache is. So guess. | |
1743 | */ | |
1744 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
1745 | if (batch * PAGE_SIZE > 512 * 1024) |
1746 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
1747 | batch /= 4; /* We effectively *= 4 below */ |
1748 | if (batch < 1) | |
1749 | batch = 1; | |
1750 | ||
1751 | /* | |
0ceaacc9 NP |
1752 | * Clamp the batch to a 2^n - 1 value. Having a power |
1753 | * of 2 value was found to be more likely to have | |
1754 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 1755 | * |
0ceaacc9 NP |
1756 | * For example if 2 tasks are alternately allocating |
1757 | * batches of pages, one task can end up with a lot | |
1758 | * of pages of one half of the possible page colors | |
1759 | * and the other with pages of the other colors. | |
e7c8d5c9 | 1760 | */ |
0ceaacc9 | 1761 | batch = (1 << (fls(batch + batch/2)-1)) - 1; |
ba56e91c | 1762 | |
e7c8d5c9 CL |
1763 | return batch; |
1764 | } | |
1765 | ||
2caaad41 CL |
1766 | inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
1767 | { | |
1768 | struct per_cpu_pages *pcp; | |
1769 | ||
1c6fe946 MD |
1770 | memset(p, 0, sizeof(*p)); |
1771 | ||
2caaad41 CL |
1772 | pcp = &p->pcp[0]; /* hot */ |
1773 | pcp->count = 0; | |
2caaad41 CL |
1774 | pcp->high = 6 * batch; |
1775 | pcp->batch = max(1UL, 1 * batch); | |
1776 | INIT_LIST_HEAD(&pcp->list); | |
1777 | ||
1778 | pcp = &p->pcp[1]; /* cold*/ | |
1779 | pcp->count = 0; | |
2caaad41 | 1780 | pcp->high = 2 * batch; |
e46a5e28 | 1781 | pcp->batch = max(1UL, batch/2); |
2caaad41 CL |
1782 | INIT_LIST_HEAD(&pcp->list); |
1783 | } | |
1784 | ||
8ad4b1fb RS |
1785 | /* |
1786 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
1787 | * to the value high for the pageset p. | |
1788 | */ | |
1789 | ||
1790 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
1791 | unsigned long high) | |
1792 | { | |
1793 | struct per_cpu_pages *pcp; | |
1794 | ||
1795 | pcp = &p->pcp[0]; /* hot list */ | |
1796 | pcp->high = high; | |
1797 | pcp->batch = max(1UL, high/4); | |
1798 | if ((high/4) > (PAGE_SHIFT * 8)) | |
1799 | pcp->batch = PAGE_SHIFT * 8; | |
1800 | } | |
1801 | ||
1802 | ||
e7c8d5c9 CL |
1803 | #ifdef CONFIG_NUMA |
1804 | /* | |
2caaad41 CL |
1805 | * Boot pageset table. One per cpu which is going to be used for all |
1806 | * zones and all nodes. The parameters will be set in such a way | |
1807 | * that an item put on a list will immediately be handed over to | |
1808 | * the buddy list. This is safe since pageset manipulation is done | |
1809 | * with interrupts disabled. | |
1810 | * | |
1811 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
b7c84c6a CL |
1812 | * |
1813 | * The boot_pagesets must be kept even after bootup is complete for | |
1814 | * unused processors and/or zones. They do play a role for bootstrapping | |
1815 | * hotplugged processors. | |
1816 | * | |
1817 | * zoneinfo_show() and maybe other functions do | |
1818 | * not check if the processor is online before following the pageset pointer. | |
1819 | * Other parts of the kernel may not check if the zone is available. | |
2caaad41 | 1820 | */ |
88a2a4ac | 1821 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; |
2caaad41 CL |
1822 | |
1823 | /* | |
1824 | * Dynamically allocate memory for the | |
e7c8d5c9 CL |
1825 | * per cpu pageset array in struct zone. |
1826 | */ | |
6292d9aa | 1827 | static int __cpuinit process_zones(int cpu) |
e7c8d5c9 CL |
1828 | { |
1829 | struct zone *zone, *dzone; | |
e7c8d5c9 CL |
1830 | |
1831 | for_each_zone(zone) { | |
e7c8d5c9 | 1832 | |
66a55030 CL |
1833 | if (!populated_zone(zone)) |
1834 | continue; | |
1835 | ||
23316bc8 | 1836 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), |
e7c8d5c9 | 1837 | GFP_KERNEL, cpu_to_node(cpu)); |
23316bc8 | 1838 | if (!zone_pcp(zone, cpu)) |
e7c8d5c9 | 1839 | goto bad; |
e7c8d5c9 | 1840 | |
23316bc8 | 1841 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); |
8ad4b1fb RS |
1842 | |
1843 | if (percpu_pagelist_fraction) | |
1844 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
1845 | (zone->present_pages / percpu_pagelist_fraction)); | |
e7c8d5c9 CL |
1846 | } |
1847 | ||
1848 | return 0; | |
1849 | bad: | |
1850 | for_each_zone(dzone) { | |
1851 | if (dzone == zone) | |
1852 | break; | |
23316bc8 NP |
1853 | kfree(zone_pcp(dzone, cpu)); |
1854 | zone_pcp(dzone, cpu) = NULL; | |
e7c8d5c9 CL |
1855 | } |
1856 | return -ENOMEM; | |
1857 | } | |
1858 | ||
1859 | static inline void free_zone_pagesets(int cpu) | |
1860 | { | |
e7c8d5c9 CL |
1861 | struct zone *zone; |
1862 | ||
1863 | for_each_zone(zone) { | |
1864 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
1865 | ||
f3ef9ead DR |
1866 | /* Free per_cpu_pageset if it is slab allocated */ |
1867 | if (pset != &boot_pageset[cpu]) | |
1868 | kfree(pset); | |
e7c8d5c9 | 1869 | zone_pcp(zone, cpu) = NULL; |
e7c8d5c9 | 1870 | } |
e7c8d5c9 CL |
1871 | } |
1872 | ||
9c7b216d | 1873 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
e7c8d5c9 CL |
1874 | unsigned long action, |
1875 | void *hcpu) | |
1876 | { | |
1877 | int cpu = (long)hcpu; | |
1878 | int ret = NOTIFY_OK; | |
1879 | ||
1880 | switch (action) { | |
1881 | case CPU_UP_PREPARE: | |
1882 | if (process_zones(cpu)) | |
1883 | ret = NOTIFY_BAD; | |
1884 | break; | |
b0d41693 | 1885 | case CPU_UP_CANCELED: |
e7c8d5c9 CL |
1886 | case CPU_DEAD: |
1887 | free_zone_pagesets(cpu); | |
1888 | break; | |
e7c8d5c9 CL |
1889 | default: |
1890 | break; | |
1891 | } | |
1892 | return ret; | |
1893 | } | |
1894 | ||
74b85f37 | 1895 | static struct notifier_block __cpuinitdata pageset_notifier = |
e7c8d5c9 CL |
1896 | { &pageset_cpuup_callback, NULL, 0 }; |
1897 | ||
78d9955b | 1898 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 CL |
1899 | { |
1900 | int err; | |
1901 | ||
1902 | /* Initialize per_cpu_pageset for cpu 0. | |
1903 | * A cpuup callback will do this for every cpu | |
1904 | * as it comes online | |
1905 | */ | |
1906 | err = process_zones(smp_processor_id()); | |
1907 | BUG_ON(err); | |
1908 | register_cpu_notifier(&pageset_notifier); | |
1909 | } | |
1910 | ||
1911 | #endif | |
1912 | ||
c09b4240 | 1913 | static __meminit |
cca448fe | 1914 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
1915 | { |
1916 | int i; | |
1917 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 1918 | size_t alloc_size; |
ed8ece2e DH |
1919 | |
1920 | /* | |
1921 | * The per-page waitqueue mechanism uses hashed waitqueues | |
1922 | * per zone. | |
1923 | */ | |
02b694de YG |
1924 | zone->wait_table_hash_nr_entries = |
1925 | wait_table_hash_nr_entries(zone_size_pages); | |
1926 | zone->wait_table_bits = | |
1927 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
1928 | alloc_size = zone->wait_table_hash_nr_entries |
1929 | * sizeof(wait_queue_head_t); | |
1930 | ||
1931 | if (system_state == SYSTEM_BOOTING) { | |
1932 | zone->wait_table = (wait_queue_head_t *) | |
1933 | alloc_bootmem_node(pgdat, alloc_size); | |
1934 | } else { | |
1935 | /* | |
1936 | * This case means that a zone whose size was 0 gets new memory | |
1937 | * via memory hot-add. | |
1938 | * But it may be the case that a new node was hot-added. In | |
1939 | * this case vmalloc() will not be able to use this new node's | |
1940 | * memory - this wait_table must be initialized to use this new | |
1941 | * node itself as well. | |
1942 | * To use this new node's memory, further consideration will be | |
1943 | * necessary. | |
1944 | */ | |
1945 | zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size); | |
1946 | } | |
1947 | if (!zone->wait_table) | |
1948 | return -ENOMEM; | |
ed8ece2e | 1949 | |
02b694de | 1950 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 1951 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
1952 | |
1953 | return 0; | |
ed8ece2e DH |
1954 | } |
1955 | ||
c09b4240 | 1956 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e DH |
1957 | { |
1958 | int cpu; | |
1959 | unsigned long batch = zone_batchsize(zone); | |
1960 | ||
1961 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
1962 | #ifdef CONFIG_NUMA | |
1963 | /* Early boot. Slab allocator not functional yet */ | |
23316bc8 | 1964 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
ed8ece2e DH |
1965 | setup_pageset(&boot_pageset[cpu],0); |
1966 | #else | |
1967 | setup_pageset(zone_pcp(zone,cpu), batch); | |
1968 | #endif | |
1969 | } | |
f5335c0f AB |
1970 | if (zone->present_pages) |
1971 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
1972 | zone->name, zone->present_pages, batch); | |
ed8ece2e DH |
1973 | } |
1974 | ||
718127cc YG |
1975 | __meminit int init_currently_empty_zone(struct zone *zone, |
1976 | unsigned long zone_start_pfn, | |
1977 | unsigned long size) | |
ed8ece2e DH |
1978 | { |
1979 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
1980 | int ret; |
1981 | ret = zone_wait_table_init(zone, size); | |
1982 | if (ret) | |
1983 | return ret; | |
ed8ece2e DH |
1984 | pgdat->nr_zones = zone_idx(zone) + 1; |
1985 | ||
ed8ece2e DH |
1986 | zone->zone_start_pfn = zone_start_pfn; |
1987 | ||
1988 | memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn); | |
1989 | ||
1990 | zone_init_free_lists(pgdat, zone, zone->spanned_pages); | |
718127cc YG |
1991 | |
1992 | return 0; | |
ed8ece2e DH |
1993 | } |
1994 | ||
c713216d MG |
1995 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
1996 | /* | |
1997 | * Basic iterator support. Return the first range of PFNs for a node | |
1998 | * Note: nid == MAX_NUMNODES returns first region regardless of node | |
1999 | */ | |
2000 | static int __init first_active_region_index_in_nid(int nid) | |
2001 | { | |
2002 | int i; | |
2003 | ||
2004 | for (i = 0; i < nr_nodemap_entries; i++) | |
2005 | if (nid == MAX_NUMNODES || early_node_map[i].nid == nid) | |
2006 | return i; | |
2007 | ||
2008 | return -1; | |
2009 | } | |
2010 | ||
2011 | /* | |
2012 | * Basic iterator support. Return the next active range of PFNs for a node | |
2013 | * Note: nid == MAX_NUMNODES returns next region regardles of node | |
2014 | */ | |
2015 | static int __init next_active_region_index_in_nid(int index, int nid) | |
2016 | { | |
2017 | for (index = index + 1; index < nr_nodemap_entries; index++) | |
2018 | if (nid == MAX_NUMNODES || early_node_map[index].nid == nid) | |
2019 | return index; | |
2020 | ||
2021 | return -1; | |
2022 | } | |
2023 | ||
2024 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID | |
2025 | /* | |
2026 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
2027 | * Architectures may implement their own version but if add_active_range() | |
2028 | * was used and there are no special requirements, this is a convenient | |
2029 | * alternative | |
2030 | */ | |
2031 | int __init early_pfn_to_nid(unsigned long pfn) | |
2032 | { | |
2033 | int i; | |
2034 | ||
2035 | for (i = 0; i < nr_nodemap_entries; i++) { | |
2036 | unsigned long start_pfn = early_node_map[i].start_pfn; | |
2037 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2038 | ||
2039 | if (start_pfn <= pfn && pfn < end_pfn) | |
2040 | return early_node_map[i].nid; | |
2041 | } | |
2042 | ||
2043 | return 0; | |
2044 | } | |
2045 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
2046 | ||
2047 | /* Basic iterator support to walk early_node_map[] */ | |
2048 | #define for_each_active_range_index_in_nid(i, nid) \ | |
2049 | for (i = first_active_region_index_in_nid(nid); i != -1; \ | |
2050 | i = next_active_region_index_in_nid(i, nid)) | |
2051 | ||
2052 | /** | |
2053 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
88ca3b94 RD |
2054 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
2055 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
c713216d MG |
2056 | * |
2057 | * If an architecture guarantees that all ranges registered with | |
2058 | * add_active_ranges() contain no holes and may be freed, this | |
2059 | * this function may be used instead of calling free_bootmem() manually. | |
2060 | */ | |
2061 | void __init free_bootmem_with_active_regions(int nid, | |
2062 | unsigned long max_low_pfn) | |
2063 | { | |
2064 | int i; | |
2065 | ||
2066 | for_each_active_range_index_in_nid(i, nid) { | |
2067 | unsigned long size_pages = 0; | |
2068 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2069 | ||
2070 | if (early_node_map[i].start_pfn >= max_low_pfn) | |
2071 | continue; | |
2072 | ||
2073 | if (end_pfn > max_low_pfn) | |
2074 | end_pfn = max_low_pfn; | |
2075 | ||
2076 | size_pages = end_pfn - early_node_map[i].start_pfn; | |
2077 | free_bootmem_node(NODE_DATA(early_node_map[i].nid), | |
2078 | PFN_PHYS(early_node_map[i].start_pfn), | |
2079 | size_pages << PAGE_SHIFT); | |
2080 | } | |
2081 | } | |
2082 | ||
2083 | /** | |
2084 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 2085 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
2086 | * |
2087 | * If an architecture guarantees that all ranges registered with | |
2088 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 2089 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
2090 | */ |
2091 | void __init sparse_memory_present_with_active_regions(int nid) | |
2092 | { | |
2093 | int i; | |
2094 | ||
2095 | for_each_active_range_index_in_nid(i, nid) | |
2096 | memory_present(early_node_map[i].nid, | |
2097 | early_node_map[i].start_pfn, | |
2098 | early_node_map[i].end_pfn); | |
2099 | } | |
2100 | ||
fb01439c MG |
2101 | /** |
2102 | * push_node_boundaries - Push node boundaries to at least the requested boundary | |
2103 | * @nid: The nid of the node to push the boundary for | |
2104 | * @start_pfn: The start pfn of the node | |
2105 | * @end_pfn: The end pfn of the node | |
2106 | * | |
2107 | * In reserve-based hot-add, mem_map is allocated that is unused until hotadd | |
2108 | * time. Specifically, on x86_64, SRAT will report ranges that can potentially | |
2109 | * be hotplugged even though no physical memory exists. This function allows | |
2110 | * an arch to push out the node boundaries so mem_map is allocated that can | |
2111 | * be used later. | |
2112 | */ | |
2113 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE | |
2114 | void __init push_node_boundaries(unsigned int nid, | |
2115 | unsigned long start_pfn, unsigned long end_pfn) | |
2116 | { | |
2117 | printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n", | |
2118 | nid, start_pfn, end_pfn); | |
2119 | ||
2120 | /* Initialise the boundary for this node if necessary */ | |
2121 | if (node_boundary_end_pfn[nid] == 0) | |
2122 | node_boundary_start_pfn[nid] = -1UL; | |
2123 | ||
2124 | /* Update the boundaries */ | |
2125 | if (node_boundary_start_pfn[nid] > start_pfn) | |
2126 | node_boundary_start_pfn[nid] = start_pfn; | |
2127 | if (node_boundary_end_pfn[nid] < end_pfn) | |
2128 | node_boundary_end_pfn[nid] = end_pfn; | |
2129 | } | |
2130 | ||
2131 | /* If necessary, push the node boundary out for reserve hotadd */ | |
2132 | static void __init account_node_boundary(unsigned int nid, | |
2133 | unsigned long *start_pfn, unsigned long *end_pfn) | |
2134 | { | |
2135 | printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n", | |
2136 | nid, *start_pfn, *end_pfn); | |
2137 | ||
2138 | /* Return if boundary information has not been provided */ | |
2139 | if (node_boundary_end_pfn[nid] == 0) | |
2140 | return; | |
2141 | ||
2142 | /* Check the boundaries and update if necessary */ | |
2143 | if (node_boundary_start_pfn[nid] < *start_pfn) | |
2144 | *start_pfn = node_boundary_start_pfn[nid]; | |
2145 | if (node_boundary_end_pfn[nid] > *end_pfn) | |
2146 | *end_pfn = node_boundary_end_pfn[nid]; | |
2147 | } | |
2148 | #else | |
2149 | void __init push_node_boundaries(unsigned int nid, | |
2150 | unsigned long start_pfn, unsigned long end_pfn) {} | |
2151 | ||
2152 | static void __init account_node_boundary(unsigned int nid, | |
2153 | unsigned long *start_pfn, unsigned long *end_pfn) {} | |
2154 | #endif | |
2155 | ||
2156 | ||
c713216d MG |
2157 | /** |
2158 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
2159 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
2160 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
2161 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
2162 | * |
2163 | * It returns the start and end page frame of a node based on information | |
2164 | * provided by an arch calling add_active_range(). If called for a node | |
2165 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 2166 | * PFNs will be 0. |
c713216d MG |
2167 | */ |
2168 | void __init get_pfn_range_for_nid(unsigned int nid, | |
2169 | unsigned long *start_pfn, unsigned long *end_pfn) | |
2170 | { | |
2171 | int i; | |
2172 | *start_pfn = -1UL; | |
2173 | *end_pfn = 0; | |
2174 | ||
2175 | for_each_active_range_index_in_nid(i, nid) { | |
2176 | *start_pfn = min(*start_pfn, early_node_map[i].start_pfn); | |
2177 | *end_pfn = max(*end_pfn, early_node_map[i].end_pfn); | |
2178 | } | |
2179 | ||
2180 | if (*start_pfn == -1UL) { | |
2181 | printk(KERN_WARNING "Node %u active with no memory\n", nid); | |
2182 | *start_pfn = 0; | |
2183 | } | |
fb01439c MG |
2184 | |
2185 | /* Push the node boundaries out if requested */ | |
2186 | account_node_boundary(nid, start_pfn, end_pfn); | |
c713216d MG |
2187 | } |
2188 | ||
2189 | /* | |
2190 | * Return the number of pages a zone spans in a node, including holes | |
2191 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
2192 | */ | |
2193 | unsigned long __init zone_spanned_pages_in_node(int nid, | |
2194 | unsigned long zone_type, | |
2195 | unsigned long *ignored) | |
2196 | { | |
2197 | unsigned long node_start_pfn, node_end_pfn; | |
2198 | unsigned long zone_start_pfn, zone_end_pfn; | |
2199 | ||
2200 | /* Get the start and end of the node and zone */ | |
2201 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
2202 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
2203 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2204 | ||
2205 | /* Check that this node has pages within the zone's required range */ | |
2206 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
2207 | return 0; | |
2208 | ||
2209 | /* Move the zone boundaries inside the node if necessary */ | |
2210 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
2211 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
2212 | ||
2213 | /* Return the spanned pages */ | |
2214 | return zone_end_pfn - zone_start_pfn; | |
2215 | } | |
2216 | ||
2217 | /* | |
2218 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 2219 | * then all holes in the requested range will be accounted for. |
c713216d MG |
2220 | */ |
2221 | unsigned long __init __absent_pages_in_range(int nid, | |
2222 | unsigned long range_start_pfn, | |
2223 | unsigned long range_end_pfn) | |
2224 | { | |
2225 | int i = 0; | |
2226 | unsigned long prev_end_pfn = 0, hole_pages = 0; | |
2227 | unsigned long start_pfn; | |
2228 | ||
2229 | /* Find the end_pfn of the first active range of pfns in the node */ | |
2230 | i = first_active_region_index_in_nid(nid); | |
2231 | if (i == -1) | |
2232 | return 0; | |
2233 | ||
9c7cd687 MG |
2234 | /* Account for ranges before physical memory on this node */ |
2235 | if (early_node_map[i].start_pfn > range_start_pfn) | |
2236 | hole_pages = early_node_map[i].start_pfn - range_start_pfn; | |
2237 | ||
c713216d MG |
2238 | prev_end_pfn = early_node_map[i].start_pfn; |
2239 | ||
2240 | /* Find all holes for the zone within the node */ | |
2241 | for (; i != -1; i = next_active_region_index_in_nid(i, nid)) { | |
2242 | ||
2243 | /* No need to continue if prev_end_pfn is outside the zone */ | |
2244 | if (prev_end_pfn >= range_end_pfn) | |
2245 | break; | |
2246 | ||
2247 | /* Make sure the end of the zone is not within the hole */ | |
2248 | start_pfn = min(early_node_map[i].start_pfn, range_end_pfn); | |
2249 | prev_end_pfn = max(prev_end_pfn, range_start_pfn); | |
2250 | ||
2251 | /* Update the hole size cound and move on */ | |
2252 | if (start_pfn > range_start_pfn) { | |
2253 | BUG_ON(prev_end_pfn > start_pfn); | |
2254 | hole_pages += start_pfn - prev_end_pfn; | |
2255 | } | |
2256 | prev_end_pfn = early_node_map[i].end_pfn; | |
2257 | } | |
2258 | ||
9c7cd687 MG |
2259 | /* Account for ranges past physical memory on this node */ |
2260 | if (range_end_pfn > prev_end_pfn) | |
2261 | hole_pages = range_end_pfn - | |
2262 | max(range_start_pfn, prev_end_pfn); | |
2263 | ||
c713216d MG |
2264 | return hole_pages; |
2265 | } | |
2266 | ||
2267 | /** | |
2268 | * absent_pages_in_range - Return number of page frames in holes within a range | |
2269 | * @start_pfn: The start PFN to start searching for holes | |
2270 | * @end_pfn: The end PFN to stop searching for holes | |
2271 | * | |
88ca3b94 | 2272 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
2273 | */ |
2274 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
2275 | unsigned long end_pfn) | |
2276 | { | |
2277 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
2278 | } | |
2279 | ||
2280 | /* Return the number of page frames in holes in a zone on a node */ | |
2281 | unsigned long __init zone_absent_pages_in_node(int nid, | |
2282 | unsigned long zone_type, | |
2283 | unsigned long *ignored) | |
2284 | { | |
9c7cd687 MG |
2285 | unsigned long node_start_pfn, node_end_pfn; |
2286 | unsigned long zone_start_pfn, zone_end_pfn; | |
2287 | ||
2288 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
2289 | zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type], | |
2290 | node_start_pfn); | |
2291 | zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type], | |
2292 | node_end_pfn); | |
2293 | ||
2294 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); | |
c713216d | 2295 | } |
0e0b864e | 2296 | |
c713216d MG |
2297 | #else |
2298 | static inline unsigned long zone_spanned_pages_in_node(int nid, | |
2299 | unsigned long zone_type, | |
2300 | unsigned long *zones_size) | |
2301 | { | |
2302 | return zones_size[zone_type]; | |
2303 | } | |
2304 | ||
2305 | static inline unsigned long zone_absent_pages_in_node(int nid, | |
2306 | unsigned long zone_type, | |
2307 | unsigned long *zholes_size) | |
2308 | { | |
2309 | if (!zholes_size) | |
2310 | return 0; | |
2311 | ||
2312 | return zholes_size[zone_type]; | |
2313 | } | |
0e0b864e | 2314 | |
c713216d MG |
2315 | #endif |
2316 | ||
2317 | static void __init calculate_node_totalpages(struct pglist_data *pgdat, | |
2318 | unsigned long *zones_size, unsigned long *zholes_size) | |
2319 | { | |
2320 | unsigned long realtotalpages, totalpages = 0; | |
2321 | enum zone_type i; | |
2322 | ||
2323 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2324 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
2325 | zones_size); | |
2326 | pgdat->node_spanned_pages = totalpages; | |
2327 | ||
2328 | realtotalpages = totalpages; | |
2329 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2330 | realtotalpages -= | |
2331 | zone_absent_pages_in_node(pgdat->node_id, i, | |
2332 | zholes_size); | |
2333 | pgdat->node_present_pages = realtotalpages; | |
2334 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
2335 | realtotalpages); | |
2336 | } | |
2337 | ||
1da177e4 LT |
2338 | /* |
2339 | * Set up the zone data structures: | |
2340 | * - mark all pages reserved | |
2341 | * - mark all memory queues empty | |
2342 | * - clear the memory bitmaps | |
2343 | */ | |
86356ab1 | 2344 | static void __meminit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
2345 | unsigned long *zones_size, unsigned long *zholes_size) |
2346 | { | |
2f1b6248 | 2347 | enum zone_type j; |
ed8ece2e | 2348 | int nid = pgdat->node_id; |
1da177e4 | 2349 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 2350 | int ret; |
1da177e4 | 2351 | |
208d54e5 | 2352 | pgdat_resize_init(pgdat); |
1da177e4 LT |
2353 | pgdat->nr_zones = 0; |
2354 | init_waitqueue_head(&pgdat->kswapd_wait); | |
2355 | pgdat->kswapd_max_order = 0; | |
2356 | ||
2357 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
2358 | struct zone *zone = pgdat->node_zones + j; | |
0e0b864e | 2359 | unsigned long size, realsize, memmap_pages; |
1da177e4 | 2360 | |
c713216d MG |
2361 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
2362 | realsize = size - zone_absent_pages_in_node(nid, j, | |
2363 | zholes_size); | |
1da177e4 | 2364 | |
0e0b864e MG |
2365 | /* |
2366 | * Adjust realsize so that it accounts for how much memory | |
2367 | * is used by this zone for memmap. This affects the watermark | |
2368 | * and per-cpu initialisations | |
2369 | */ | |
2370 | memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT; | |
2371 | if (realsize >= memmap_pages) { | |
2372 | realsize -= memmap_pages; | |
2373 | printk(KERN_DEBUG | |
2374 | " %s zone: %lu pages used for memmap\n", | |
2375 | zone_names[j], memmap_pages); | |
2376 | } else | |
2377 | printk(KERN_WARNING | |
2378 | " %s zone: %lu pages exceeds realsize %lu\n", | |
2379 | zone_names[j], memmap_pages, realsize); | |
2380 | ||
2381 | /* Account for reserved DMA pages */ | |
2382 | if (j == ZONE_DMA && realsize > dma_reserve) { | |
2383 | realsize -= dma_reserve; | |
2384 | printk(KERN_DEBUG " DMA zone: %lu pages reserved\n", | |
2385 | dma_reserve); | |
2386 | } | |
2387 | ||
98d2b0eb | 2388 | if (!is_highmem_idx(j)) |
1da177e4 LT |
2389 | nr_kernel_pages += realsize; |
2390 | nr_all_pages += realsize; | |
2391 | ||
2392 | zone->spanned_pages = size; | |
2393 | zone->present_pages = realsize; | |
9614634f | 2394 | #ifdef CONFIG_NUMA |
d5f541ed | 2395 | zone->node = nid; |
8417bba4 | 2396 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
9614634f | 2397 | / 100; |
0ff38490 | 2398 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
9614634f | 2399 | #endif |
1da177e4 LT |
2400 | zone->name = zone_names[j]; |
2401 | spin_lock_init(&zone->lock); | |
2402 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 2403 | zone_seqlock_init(zone); |
1da177e4 LT |
2404 | zone->zone_pgdat = pgdat; |
2405 | zone->free_pages = 0; | |
2406 | ||
2407 | zone->temp_priority = zone->prev_priority = DEF_PRIORITY; | |
2408 | ||
ed8ece2e | 2409 | zone_pcp_init(zone); |
1da177e4 LT |
2410 | INIT_LIST_HEAD(&zone->active_list); |
2411 | INIT_LIST_HEAD(&zone->inactive_list); | |
2412 | zone->nr_scan_active = 0; | |
2413 | zone->nr_scan_inactive = 0; | |
2414 | zone->nr_active = 0; | |
2415 | zone->nr_inactive = 0; | |
2244b95a | 2416 | zap_zone_vm_stats(zone); |
53e9a615 | 2417 | atomic_set(&zone->reclaim_in_progress, 0); |
1da177e4 LT |
2418 | if (!size) |
2419 | continue; | |
2420 | ||
d41dee36 | 2421 | zonetable_add(zone, nid, j, zone_start_pfn, size); |
718127cc YG |
2422 | ret = init_currently_empty_zone(zone, zone_start_pfn, size); |
2423 | BUG_ON(ret); | |
1da177e4 | 2424 | zone_start_pfn += size; |
1da177e4 LT |
2425 | } |
2426 | } | |
2427 | ||
2428 | static void __init alloc_node_mem_map(struct pglist_data *pgdat) | |
2429 | { | |
1da177e4 LT |
2430 | /* Skip empty nodes */ |
2431 | if (!pgdat->node_spanned_pages) | |
2432 | return; | |
2433 | ||
d41dee36 | 2434 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
2435 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
2436 | if (!pgdat->node_mem_map) { | |
e984bb43 | 2437 | unsigned long size, start, end; |
d41dee36 AW |
2438 | struct page *map; |
2439 | ||
e984bb43 BP |
2440 | /* |
2441 | * The zone's endpoints aren't required to be MAX_ORDER | |
2442 | * aligned but the node_mem_map endpoints must be in order | |
2443 | * for the buddy allocator to function correctly. | |
2444 | */ | |
2445 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
2446 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
2447 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
2448 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
2449 | map = alloc_remap(pgdat->node_id, size); |
2450 | if (!map) | |
2451 | map = alloc_bootmem_node(pgdat, size); | |
e984bb43 | 2452 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 2453 | } |
d41dee36 | 2454 | #ifdef CONFIG_FLATMEM |
1da177e4 LT |
2455 | /* |
2456 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
2457 | */ | |
c713216d | 2458 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 2459 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d MG |
2460 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
2461 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) | |
2462 | mem_map -= pgdat->node_start_pfn; | |
2463 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
2464 | } | |
1da177e4 | 2465 | #endif |
d41dee36 | 2466 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
2467 | } |
2468 | ||
86356ab1 | 2469 | void __meminit free_area_init_node(int nid, struct pglist_data *pgdat, |
1da177e4 LT |
2470 | unsigned long *zones_size, unsigned long node_start_pfn, |
2471 | unsigned long *zholes_size) | |
2472 | { | |
2473 | pgdat->node_id = nid; | |
2474 | pgdat->node_start_pfn = node_start_pfn; | |
c713216d | 2475 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
1da177e4 LT |
2476 | |
2477 | alloc_node_mem_map(pgdat); | |
2478 | ||
2479 | free_area_init_core(pgdat, zones_size, zholes_size); | |
2480 | } | |
2481 | ||
c713216d MG |
2482 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
2483 | /** | |
2484 | * add_active_range - Register a range of PFNs backed by physical memory | |
2485 | * @nid: The node ID the range resides on | |
2486 | * @start_pfn: The start PFN of the available physical memory | |
2487 | * @end_pfn: The end PFN of the available physical memory | |
2488 | * | |
2489 | * These ranges are stored in an early_node_map[] and later used by | |
2490 | * free_area_init_nodes() to calculate zone sizes and holes. If the | |
2491 | * range spans a memory hole, it is up to the architecture to ensure | |
2492 | * the memory is not freed by the bootmem allocator. If possible | |
2493 | * the range being registered will be merged with existing ranges. | |
2494 | */ | |
2495 | void __init add_active_range(unsigned int nid, unsigned long start_pfn, | |
2496 | unsigned long end_pfn) | |
2497 | { | |
2498 | int i; | |
2499 | ||
2500 | printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) " | |
2501 | "%d entries of %d used\n", | |
2502 | nid, start_pfn, end_pfn, | |
2503 | nr_nodemap_entries, MAX_ACTIVE_REGIONS); | |
2504 | ||
2505 | /* Merge with existing active regions if possible */ | |
2506 | for (i = 0; i < nr_nodemap_entries; i++) { | |
2507 | if (early_node_map[i].nid != nid) | |
2508 | continue; | |
2509 | ||
2510 | /* Skip if an existing region covers this new one */ | |
2511 | if (start_pfn >= early_node_map[i].start_pfn && | |
2512 | end_pfn <= early_node_map[i].end_pfn) | |
2513 | return; | |
2514 | ||
2515 | /* Merge forward if suitable */ | |
2516 | if (start_pfn <= early_node_map[i].end_pfn && | |
2517 | end_pfn > early_node_map[i].end_pfn) { | |
2518 | early_node_map[i].end_pfn = end_pfn; | |
2519 | return; | |
2520 | } | |
2521 | ||
2522 | /* Merge backward if suitable */ | |
2523 | if (start_pfn < early_node_map[i].end_pfn && | |
2524 | end_pfn >= early_node_map[i].start_pfn) { | |
2525 | early_node_map[i].start_pfn = start_pfn; | |
2526 | return; | |
2527 | } | |
2528 | } | |
2529 | ||
2530 | /* Check that early_node_map is large enough */ | |
2531 | if (i >= MAX_ACTIVE_REGIONS) { | |
2532 | printk(KERN_CRIT "More than %d memory regions, truncating\n", | |
2533 | MAX_ACTIVE_REGIONS); | |
2534 | return; | |
2535 | } | |
2536 | ||
2537 | early_node_map[i].nid = nid; | |
2538 | early_node_map[i].start_pfn = start_pfn; | |
2539 | early_node_map[i].end_pfn = end_pfn; | |
2540 | nr_nodemap_entries = i + 1; | |
2541 | } | |
2542 | ||
2543 | /** | |
2544 | * shrink_active_range - Shrink an existing registered range of PFNs | |
2545 | * @nid: The node id the range is on that should be shrunk | |
2546 | * @old_end_pfn: The old end PFN of the range | |
2547 | * @new_end_pfn: The new PFN of the range | |
2548 | * | |
2549 | * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node. | |
2550 | * The map is kept at the end physical page range that has already been | |
2551 | * registered with add_active_range(). This function allows an arch to shrink | |
2552 | * an existing registered range. | |
2553 | */ | |
2554 | void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn, | |
2555 | unsigned long new_end_pfn) | |
2556 | { | |
2557 | int i; | |
2558 | ||
2559 | /* Find the old active region end and shrink */ | |
2560 | for_each_active_range_index_in_nid(i, nid) | |
2561 | if (early_node_map[i].end_pfn == old_end_pfn) { | |
2562 | early_node_map[i].end_pfn = new_end_pfn; | |
2563 | break; | |
2564 | } | |
2565 | } | |
2566 | ||
2567 | /** | |
2568 | * remove_all_active_ranges - Remove all currently registered regions | |
88ca3b94 | 2569 | * |
c713216d MG |
2570 | * During discovery, it may be found that a table like SRAT is invalid |
2571 | * and an alternative discovery method must be used. This function removes | |
2572 | * all currently registered regions. | |
2573 | */ | |
88ca3b94 | 2574 | void __init remove_all_active_ranges(void) |
c713216d MG |
2575 | { |
2576 | memset(early_node_map, 0, sizeof(early_node_map)); | |
2577 | nr_nodemap_entries = 0; | |
fb01439c MG |
2578 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE |
2579 | memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn)); | |
2580 | memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn)); | |
2581 | #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */ | |
c713216d MG |
2582 | } |
2583 | ||
2584 | /* Compare two active node_active_regions */ | |
2585 | static int __init cmp_node_active_region(const void *a, const void *b) | |
2586 | { | |
2587 | struct node_active_region *arange = (struct node_active_region *)a; | |
2588 | struct node_active_region *brange = (struct node_active_region *)b; | |
2589 | ||
2590 | /* Done this way to avoid overflows */ | |
2591 | if (arange->start_pfn > brange->start_pfn) | |
2592 | return 1; | |
2593 | if (arange->start_pfn < brange->start_pfn) | |
2594 | return -1; | |
2595 | ||
2596 | return 0; | |
2597 | } | |
2598 | ||
2599 | /* sort the node_map by start_pfn */ | |
2600 | static void __init sort_node_map(void) | |
2601 | { | |
2602 | sort(early_node_map, (size_t)nr_nodemap_entries, | |
2603 | sizeof(struct node_active_region), | |
2604 | cmp_node_active_region, NULL); | |
2605 | } | |
2606 | ||
2607 | /* Find the lowest pfn for a node. This depends on a sorted early_node_map */ | |
2608 | unsigned long __init find_min_pfn_for_node(unsigned long nid) | |
2609 | { | |
2610 | int i; | |
2611 | ||
2612 | /* Assuming a sorted map, the first range found has the starting pfn */ | |
2613 | for_each_active_range_index_in_nid(i, nid) | |
2614 | return early_node_map[i].start_pfn; | |
2615 | ||
2616 | printk(KERN_WARNING "Could not find start_pfn for node %lu\n", nid); | |
2617 | return 0; | |
2618 | } | |
2619 | ||
2620 | /** | |
2621 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
2622 | * | |
2623 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 2624 | * add_active_range(). |
c713216d MG |
2625 | */ |
2626 | unsigned long __init find_min_pfn_with_active_regions(void) | |
2627 | { | |
2628 | return find_min_pfn_for_node(MAX_NUMNODES); | |
2629 | } | |
2630 | ||
2631 | /** | |
2632 | * find_max_pfn_with_active_regions - Find the maximum PFN registered | |
2633 | * | |
2634 | * It returns the maximum PFN based on information provided via | |
88ca3b94 | 2635 | * add_active_range(). |
c713216d MG |
2636 | */ |
2637 | unsigned long __init find_max_pfn_with_active_regions(void) | |
2638 | { | |
2639 | int i; | |
2640 | unsigned long max_pfn = 0; | |
2641 | ||
2642 | for (i = 0; i < nr_nodemap_entries; i++) | |
2643 | max_pfn = max(max_pfn, early_node_map[i].end_pfn); | |
2644 | ||
2645 | return max_pfn; | |
2646 | } | |
2647 | ||
2648 | /** | |
2649 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 2650 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
2651 | * |
2652 | * This will call free_area_init_node() for each active node in the system. | |
2653 | * Using the page ranges provided by add_active_range(), the size of each | |
2654 | * zone in each node and their holes is calculated. If the maximum PFN | |
2655 | * between two adjacent zones match, it is assumed that the zone is empty. | |
2656 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
2657 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
2658 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
2659 | * at arch_max_dma_pfn. | |
2660 | */ | |
2661 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
2662 | { | |
2663 | unsigned long nid; | |
2664 | enum zone_type i; | |
2665 | ||
2666 | /* Record where the zone boundaries are */ | |
2667 | memset(arch_zone_lowest_possible_pfn, 0, | |
2668 | sizeof(arch_zone_lowest_possible_pfn)); | |
2669 | memset(arch_zone_highest_possible_pfn, 0, | |
2670 | sizeof(arch_zone_highest_possible_pfn)); | |
2671 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
2672 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
2673 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2674 | arch_zone_lowest_possible_pfn[i] = | |
2675 | arch_zone_highest_possible_pfn[i-1]; | |
2676 | arch_zone_highest_possible_pfn[i] = | |
2677 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
2678 | } | |
2679 | ||
2680 | /* Regions in the early_node_map can be in any order */ | |
2681 | sort_node_map(); | |
2682 | ||
2683 | /* Print out the zone ranges */ | |
2684 | printk("Zone PFN ranges:\n"); | |
2685 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2686 | printk(" %-8s %8lu -> %8lu\n", | |
2687 | zone_names[i], | |
2688 | arch_zone_lowest_possible_pfn[i], | |
2689 | arch_zone_highest_possible_pfn[i]); | |
2690 | ||
2691 | /* Print out the early_node_map[] */ | |
2692 | printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries); | |
2693 | for (i = 0; i < nr_nodemap_entries; i++) | |
2694 | printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid, | |
2695 | early_node_map[i].start_pfn, | |
2696 | early_node_map[i].end_pfn); | |
2697 | ||
2698 | /* Initialise every node */ | |
2699 | for_each_online_node(nid) { | |
2700 | pg_data_t *pgdat = NODE_DATA(nid); | |
2701 | free_area_init_node(nid, pgdat, NULL, | |
2702 | find_min_pfn_for_node(nid), NULL); | |
2703 | } | |
2704 | } | |
2705 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
2706 | ||
0e0b864e | 2707 | /** |
88ca3b94 RD |
2708 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
2709 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
2710 | * |
2711 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
2712 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
2713 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
2714 | * function may optionally be used to account for unfreeable pages in the |
2715 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
2716 | * smaller per-cpu batchsize. | |
0e0b864e MG |
2717 | */ |
2718 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
2719 | { | |
2720 | dma_reserve = new_dma_reserve; | |
2721 | } | |
2722 | ||
93b7504e | 2723 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
2724 | static bootmem_data_t contig_bootmem_data; |
2725 | struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data }; | |
2726 | ||
2727 | EXPORT_SYMBOL(contig_page_data); | |
93b7504e | 2728 | #endif |
1da177e4 LT |
2729 | |
2730 | void __init free_area_init(unsigned long *zones_size) | |
2731 | { | |
93b7504e | 2732 | free_area_init_node(0, NODE_DATA(0), zones_size, |
1da177e4 LT |
2733 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
2734 | } | |
1da177e4 | 2735 | |
1da177e4 LT |
2736 | #ifdef CONFIG_HOTPLUG_CPU |
2737 | static int page_alloc_cpu_notify(struct notifier_block *self, | |
2738 | unsigned long action, void *hcpu) | |
2739 | { | |
2740 | int cpu = (unsigned long)hcpu; | |
1da177e4 LT |
2741 | |
2742 | if (action == CPU_DEAD) { | |
1da177e4 LT |
2743 | local_irq_disable(); |
2744 | __drain_pages(cpu); | |
f8891e5e | 2745 | vm_events_fold_cpu(cpu); |
1da177e4 | 2746 | local_irq_enable(); |
2244b95a | 2747 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
2748 | } |
2749 | return NOTIFY_OK; | |
2750 | } | |
2751 | #endif /* CONFIG_HOTPLUG_CPU */ | |
2752 | ||
2753 | void __init page_alloc_init(void) | |
2754 | { | |
2755 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
2756 | } | |
2757 | ||
cb45b0e9 HA |
2758 | /* |
2759 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
2760 | * or min_free_kbytes changes. | |
2761 | */ | |
2762 | static void calculate_totalreserve_pages(void) | |
2763 | { | |
2764 | struct pglist_data *pgdat; | |
2765 | unsigned long reserve_pages = 0; | |
2f6726e5 | 2766 | enum zone_type i, j; |
cb45b0e9 HA |
2767 | |
2768 | for_each_online_pgdat(pgdat) { | |
2769 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2770 | struct zone *zone = pgdat->node_zones + i; | |
2771 | unsigned long max = 0; | |
2772 | ||
2773 | /* Find valid and maximum lowmem_reserve in the zone */ | |
2774 | for (j = i; j < MAX_NR_ZONES; j++) { | |
2775 | if (zone->lowmem_reserve[j] > max) | |
2776 | max = zone->lowmem_reserve[j]; | |
2777 | } | |
2778 | ||
2779 | /* we treat pages_high as reserved pages. */ | |
2780 | max += zone->pages_high; | |
2781 | ||
2782 | if (max > zone->present_pages) | |
2783 | max = zone->present_pages; | |
2784 | reserve_pages += max; | |
2785 | } | |
2786 | } | |
2787 | totalreserve_pages = reserve_pages; | |
2788 | } | |
2789 | ||
1da177e4 LT |
2790 | /* |
2791 | * setup_per_zone_lowmem_reserve - called whenever | |
2792 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
2793 | * has a correct pages reserved value, so an adequate number of | |
2794 | * pages are left in the zone after a successful __alloc_pages(). | |
2795 | */ | |
2796 | static void setup_per_zone_lowmem_reserve(void) | |
2797 | { | |
2798 | struct pglist_data *pgdat; | |
2f6726e5 | 2799 | enum zone_type j, idx; |
1da177e4 | 2800 | |
ec936fc5 | 2801 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
2802 | for (j = 0; j < MAX_NR_ZONES; j++) { |
2803 | struct zone *zone = pgdat->node_zones + j; | |
2804 | unsigned long present_pages = zone->present_pages; | |
2805 | ||
2806 | zone->lowmem_reserve[j] = 0; | |
2807 | ||
2f6726e5 CL |
2808 | idx = j; |
2809 | while (idx) { | |
1da177e4 LT |
2810 | struct zone *lower_zone; |
2811 | ||
2f6726e5 CL |
2812 | idx--; |
2813 | ||
1da177e4 LT |
2814 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
2815 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
2816 | ||
2817 | lower_zone = pgdat->node_zones + idx; | |
2818 | lower_zone->lowmem_reserve[j] = present_pages / | |
2819 | sysctl_lowmem_reserve_ratio[idx]; | |
2820 | present_pages += lower_zone->present_pages; | |
2821 | } | |
2822 | } | |
2823 | } | |
cb45b0e9 HA |
2824 | |
2825 | /* update totalreserve_pages */ | |
2826 | calculate_totalreserve_pages(); | |
1da177e4 LT |
2827 | } |
2828 | ||
88ca3b94 RD |
2829 | /** |
2830 | * setup_per_zone_pages_min - called when min_free_kbytes changes. | |
2831 | * | |
2832 | * Ensures that the pages_{min,low,high} values for each zone are set correctly | |
2833 | * with respect to min_free_kbytes. | |
1da177e4 | 2834 | */ |
3947be19 | 2835 | void setup_per_zone_pages_min(void) |
1da177e4 LT |
2836 | { |
2837 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
2838 | unsigned long lowmem_pages = 0; | |
2839 | struct zone *zone; | |
2840 | unsigned long flags; | |
2841 | ||
2842 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
2843 | for_each_zone(zone) { | |
2844 | if (!is_highmem(zone)) | |
2845 | lowmem_pages += zone->present_pages; | |
2846 | } | |
2847 | ||
2848 | for_each_zone(zone) { | |
ac924c60 AM |
2849 | u64 tmp; |
2850 | ||
1da177e4 | 2851 | spin_lock_irqsave(&zone->lru_lock, flags); |
ac924c60 AM |
2852 | tmp = (u64)pages_min * zone->present_pages; |
2853 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
2854 | if (is_highmem(zone)) { |
2855 | /* | |
669ed175 NP |
2856 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
2857 | * need highmem pages, so cap pages_min to a small | |
2858 | * value here. | |
2859 | * | |
2860 | * The (pages_high-pages_low) and (pages_low-pages_min) | |
2861 | * deltas controls asynch page reclaim, and so should | |
2862 | * not be capped for highmem. | |
1da177e4 LT |
2863 | */ |
2864 | int min_pages; | |
2865 | ||
2866 | min_pages = zone->present_pages / 1024; | |
2867 | if (min_pages < SWAP_CLUSTER_MAX) | |
2868 | min_pages = SWAP_CLUSTER_MAX; | |
2869 | if (min_pages > 128) | |
2870 | min_pages = 128; | |
2871 | zone->pages_min = min_pages; | |
2872 | } else { | |
669ed175 NP |
2873 | /* |
2874 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
2875 | * proportionate to the zone's size. |
2876 | */ | |
669ed175 | 2877 | zone->pages_min = tmp; |
1da177e4 LT |
2878 | } |
2879 | ||
ac924c60 AM |
2880 | zone->pages_low = zone->pages_min + (tmp >> 2); |
2881 | zone->pages_high = zone->pages_min + (tmp >> 1); | |
1da177e4 LT |
2882 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
2883 | } | |
cb45b0e9 HA |
2884 | |
2885 | /* update totalreserve_pages */ | |
2886 | calculate_totalreserve_pages(); | |
1da177e4 LT |
2887 | } |
2888 | ||
2889 | /* | |
2890 | * Initialise min_free_kbytes. | |
2891 | * | |
2892 | * For small machines we want it small (128k min). For large machines | |
2893 | * we want it large (64MB max). But it is not linear, because network | |
2894 | * bandwidth does not increase linearly with machine size. We use | |
2895 | * | |
2896 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
2897 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
2898 | * | |
2899 | * which yields | |
2900 | * | |
2901 | * 16MB: 512k | |
2902 | * 32MB: 724k | |
2903 | * 64MB: 1024k | |
2904 | * 128MB: 1448k | |
2905 | * 256MB: 2048k | |
2906 | * 512MB: 2896k | |
2907 | * 1024MB: 4096k | |
2908 | * 2048MB: 5792k | |
2909 | * 4096MB: 8192k | |
2910 | * 8192MB: 11584k | |
2911 | * 16384MB: 16384k | |
2912 | */ | |
2913 | static int __init init_per_zone_pages_min(void) | |
2914 | { | |
2915 | unsigned long lowmem_kbytes; | |
2916 | ||
2917 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
2918 | ||
2919 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
2920 | if (min_free_kbytes < 128) | |
2921 | min_free_kbytes = 128; | |
2922 | if (min_free_kbytes > 65536) | |
2923 | min_free_kbytes = 65536; | |
2924 | setup_per_zone_pages_min(); | |
2925 | setup_per_zone_lowmem_reserve(); | |
2926 | return 0; | |
2927 | } | |
2928 | module_init(init_per_zone_pages_min) | |
2929 | ||
2930 | /* | |
2931 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
2932 | * that we can call two helper functions whenever min_free_kbytes | |
2933 | * changes. | |
2934 | */ | |
2935 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
2936 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2937 | { | |
2938 | proc_dointvec(table, write, file, buffer, length, ppos); | |
2939 | setup_per_zone_pages_min(); | |
2940 | return 0; | |
2941 | } | |
2942 | ||
9614634f CL |
2943 | #ifdef CONFIG_NUMA |
2944 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
2945 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2946 | { | |
2947 | struct zone *zone; | |
2948 | int rc; | |
2949 | ||
2950 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2951 | if (rc) | |
2952 | return rc; | |
2953 | ||
2954 | for_each_zone(zone) | |
8417bba4 | 2955 | zone->min_unmapped_pages = (zone->present_pages * |
9614634f CL |
2956 | sysctl_min_unmapped_ratio) / 100; |
2957 | return 0; | |
2958 | } | |
0ff38490 CL |
2959 | |
2960 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
2961 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2962 | { | |
2963 | struct zone *zone; | |
2964 | int rc; | |
2965 | ||
2966 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2967 | if (rc) | |
2968 | return rc; | |
2969 | ||
2970 | for_each_zone(zone) | |
2971 | zone->min_slab_pages = (zone->present_pages * | |
2972 | sysctl_min_slab_ratio) / 100; | |
2973 | return 0; | |
2974 | } | |
9614634f CL |
2975 | #endif |
2976 | ||
1da177e4 LT |
2977 | /* |
2978 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
2979 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
2980 | * whenever sysctl_lowmem_reserve_ratio changes. | |
2981 | * | |
2982 | * The reserve ratio obviously has absolutely no relation with the | |
2983 | * pages_min watermarks. The lowmem reserve ratio can only make sense | |
2984 | * if in function of the boot time zone sizes. | |
2985 | */ | |
2986 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
2987 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2988 | { | |
2989 | proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2990 | setup_per_zone_lowmem_reserve(); | |
2991 | return 0; | |
2992 | } | |
2993 | ||
8ad4b1fb RS |
2994 | /* |
2995 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
2996 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
2997 | * can have before it gets flushed back to buddy allocator. | |
2998 | */ | |
2999 | ||
3000 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
3001 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3002 | { | |
3003 | struct zone *zone; | |
3004 | unsigned int cpu; | |
3005 | int ret; | |
3006 | ||
3007 | ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
3008 | if (!write || (ret == -EINVAL)) | |
3009 | return ret; | |
3010 | for_each_zone(zone) { | |
3011 | for_each_online_cpu(cpu) { | |
3012 | unsigned long high; | |
3013 | high = zone->present_pages / percpu_pagelist_fraction; | |
3014 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
3015 | } | |
3016 | } | |
3017 | return 0; | |
3018 | } | |
3019 | ||
f034b5d4 | 3020 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
3021 | |
3022 | #ifdef CONFIG_NUMA | |
3023 | static int __init set_hashdist(char *str) | |
3024 | { | |
3025 | if (!str) | |
3026 | return 0; | |
3027 | hashdist = simple_strtoul(str, &str, 0); | |
3028 | return 1; | |
3029 | } | |
3030 | __setup("hashdist=", set_hashdist); | |
3031 | #endif | |
3032 | ||
3033 | /* | |
3034 | * allocate a large system hash table from bootmem | |
3035 | * - it is assumed that the hash table must contain an exact power-of-2 | |
3036 | * quantity of entries | |
3037 | * - limit is the number of hash buckets, not the total allocation size | |
3038 | */ | |
3039 | void *__init alloc_large_system_hash(const char *tablename, | |
3040 | unsigned long bucketsize, | |
3041 | unsigned long numentries, | |
3042 | int scale, | |
3043 | int flags, | |
3044 | unsigned int *_hash_shift, | |
3045 | unsigned int *_hash_mask, | |
3046 | unsigned long limit) | |
3047 | { | |
3048 | unsigned long long max = limit; | |
3049 | unsigned long log2qty, size; | |
3050 | void *table = NULL; | |
3051 | ||
3052 | /* allow the kernel cmdline to have a say */ | |
3053 | if (!numentries) { | |
3054 | /* round applicable memory size up to nearest megabyte */ | |
3055 | numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages; | |
3056 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; | |
3057 | numentries >>= 20 - PAGE_SHIFT; | |
3058 | numentries <<= 20 - PAGE_SHIFT; | |
3059 | ||
3060 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
3061 | if (scale > PAGE_SHIFT) | |
3062 | numentries >>= (scale - PAGE_SHIFT); | |
3063 | else | |
3064 | numentries <<= (PAGE_SHIFT - scale); | |
3065 | } | |
6e692ed3 | 3066 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
3067 | |
3068 | /* limit allocation size to 1/16 total memory by default */ | |
3069 | if (max == 0) { | |
3070 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
3071 | do_div(max, bucketsize); | |
3072 | } | |
3073 | ||
3074 | if (numentries > max) | |
3075 | numentries = max; | |
3076 | ||
3077 | log2qty = long_log2(numentries); | |
3078 | ||
3079 | do { | |
3080 | size = bucketsize << log2qty; | |
3081 | if (flags & HASH_EARLY) | |
3082 | table = alloc_bootmem(size); | |
3083 | else if (hashdist) | |
3084 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
3085 | else { | |
3086 | unsigned long order; | |
3087 | for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++) | |
3088 | ; | |
3089 | table = (void*) __get_free_pages(GFP_ATOMIC, order); | |
3090 | } | |
3091 | } while (!table && size > PAGE_SIZE && --log2qty); | |
3092 | ||
3093 | if (!table) | |
3094 | panic("Failed to allocate %s hash table\n", tablename); | |
3095 | ||
3096 | printk("%s hash table entries: %d (order: %d, %lu bytes)\n", | |
3097 | tablename, | |
3098 | (1U << log2qty), | |
3099 | long_log2(size) - PAGE_SHIFT, | |
3100 | size); | |
3101 | ||
3102 | if (_hash_shift) | |
3103 | *_hash_shift = log2qty; | |
3104 | if (_hash_mask) | |
3105 | *_hash_mask = (1 << log2qty) - 1; | |
3106 | ||
3107 | return table; | |
3108 | } | |
a117e66e KH |
3109 | |
3110 | #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE | |
a117e66e KH |
3111 | struct page *pfn_to_page(unsigned long pfn) |
3112 | { | |
67de6482 | 3113 | return __pfn_to_page(pfn); |
a117e66e KH |
3114 | } |
3115 | unsigned long page_to_pfn(struct page *page) | |
3116 | { | |
67de6482 | 3117 | return __page_to_pfn(page); |
a117e66e | 3118 | } |
a117e66e KH |
3119 | EXPORT_SYMBOL(pfn_to_page); |
3120 | EXPORT_SYMBOL(page_to_pfn); | |
3121 | #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */ |